Computerworld.com [Hacking News]

IT industry experts and analysts are still trying to piece together Nvidia’s surprise plan to pay $12.9 billion for open-source AI company Hugging Face.

Nvidia dominates AI with its GPUs, and the company generates billions of dollars in revenue through a proprietary approach to the fast-moving technology. Hugging Face, on the other hand, hosts open models and has been a neutral player between chip vendors and model labs.

“This is about Nvidia having more say in how the stack gets built,” said Stephanie Walter, analyst at Hyperframe Research.

Hugging Face is wildly popular with developers, and Nvidia is buying early influence with that crowd. “You have a better chance of being part of the production environment later,” Walter said, adding that she wasn’t sure how Nvidia reached a nearly $13 billion price tag for the acquisition.

“Hugging Face has near-uncontested market primacy over where developers go for open-weight model releases. Now Nvidia owns that,” said Mark Petty, senior director analyst at Gartner.

Nvidia’s chase for developers should force IT decision-makers to review how much of the AI stack they control, said Hector Liu, director of Institute of Foundation Models’ Silicon Valley Lab. IFM is part of the Abu Dhabi-based Mohamed bin Zayed University of Artificial Intelligence.

Liu said CIOs should ask themselves three questions: “Can you run the model on hardware you already have, without a dependency you didn’t choose? Can you see how it was built?” And, “is the license one you can build a business on?

“A model that passes all three is durable, no matter who buys whom next year,” Liu said.

IFM’s latest K2 Horizon model, which was introduced on the same day Nvidia’s deal was announced, is hosted on Hugging Face. The open model was built to answer all of those questions.

K2 Horizon runs on AI hardware from Nvidia, AMD, Cerebras, and major cloud providers, said Liu, who doesn’t expect that to change. “Nvidia has said Hugging Face stays an open platform for every builder and every accelerator, and we’ll take that at face value,” Liu said.

Nvidia pledged to maintain Hugging Face’s hardware and model independence, and said its compute won’t be required. 

Even so, Nvidia isn’t paying nearly $13 billion for a model repository, said Jake Newfield, CEO at Hermetiq — it’s buying the front door to open AI.

(Hermetiq develops AI build and code observability tools.)

AI-generated output is becoming abundant and the infrastructure that makes it testable, reproducible and deployable is becoming strategically valuable, Newfield said. “Nvidia can accelerate it with capital and compute, but the real test is operational neutrality,” he said. 

That involves assessing whether competing hardware remains equally supported across the tooling, benchmarks and deployment paths developers actually use, Newfield said.

Beyond distribution, Nvidia is also buying Hugging Face for data, Petty said. That data showed that agents overtook humans as its largest traffic source in July, the kind of insight that could prove valuable down the road.

“Every model pulled tells Nvidia what the market wants next,” Petty said.

Nvidia has every reason to grow demand for open models and to keep them cheap, Petty said. “That’s good for enterprises, as it prevents market consolidation around a small number of proprietary model owners,” Petty said.

Closed and open models will coexist and lead to a world “where you’re going to continue to train these models and run these models at scale,” Justin Boitano, Nvidia’s vice president for Enterprise AI, said in a press conference after the deal’s announcement.

Nvidia will benefit “through the training that’s done and the inference that’s done on our hardware as models get diffused into the ecosystem at scale,” Boitano said.

That’s one motivation to keep the ecosystem open and neutral, so “developers can work wherever they want to work,” Boitano said.

Open source makes AI more accessible by lowering the barriers to experimentation and adoption, said Jon Carvill, senior vice president of marketing at AI chip maker Nuvacore. “Bringing Nvidia and Hugging Face closer together should help accelerate that choice, access and innovation,” Carvill said.

But there are still unanswered questions around whether Hugging Face will remain open and how much proprietary control Nvidia might exert, said Jack Gold, principal analyst at J. Gold Associates.

Microsoft traveled a similar path with its acquisition of open-source repository GitHub, which “did not really pan out as well as Microsoft hoped,” Gold said. “With Nvidia’s acquisition, will it still be as open to competitive hardware-software access, or will there be some barriers employed?”

I surprised myself this morning when I came across an interesting Apple-commissioned report — Rethinking critical AI infrastructure — I’d not seen before. It looks at the shifting expectations for AI infrastructure and recognizes that enterprise users want (and need) secure, on-device AI solutions for critical parts of their business.

That’s why tens of thousands of companies are already investing in Macs, because they recognize that Macs do indeed do AI. The study, published earlier this year and put together by Omdia, reflects insights gathered across 1,500 conversations with enterprise tech leaders and practitioners, noting that for many in business the current cloud-based approach to AI fails to deliver on three key metrics:

• Costs: Current pricing models seem unsustainable. Particularly when it comes to agentic AI, costs climb fast and business users need to get those costs under control. 
• Security: Even the most secure cloud services include some degree of data risk. When it comes to using AI for regulated data in industries such as healthcare, business users need much more security than the cloud inherently provides. After all, data that is not transmitted will not leak in transmission.
• >Capacity>: Workload requirements change and capacity needs to scale. That can boost the cost of accessing additional cloud capacity, or impose limitations in the event it can’t be found. It’s also true that while frontier models can provide all the bells and whistles of AI for advanced tasks, the vast majority of the AI work does not require anything near as much power. As Omdia explains: “57% of enterprise models are under 10 billion parameters, well within the capabilities of modern devices like MacBook Air or the entry-level MacBook Pro.”

What they think

As you might expect, the researchers believe on-premises AI set-ups respond to all three needs; not only that, but once you’ve coughed up cash for the necessary computational infrastructure, you don’t have to pay much more. “On-device infrastructure has near-zero marginal cost after initial investment, enabling unlimited experimentation without budget constraints,” the report said. 

Basically, once you’ve invested in on-premises capacity, you can divert mundane AI tasks to those machines for processing — limiting costs, boosting security and releasing capacity, turning to cloud-based models only when higher end AI solutions are required. While that’s good news for Apple, that’s bad news for many AI companies’ revenue models. (Perhaps they should have recognized that even the most advanced LLM’s will run on a standard iPhone eventually.)

The other advantage is that if AI is not used as widely as expected across a company, the same hardware can be used for other company tasks.

What’s actually happening

Enterprises already using AI are learning these lessons, which is why we see more of them buying Macs for these tasks. They do so because Apple’s computers deliver the computational power and performance to run AI effectively, from chip design to power consumption to the OS itself. Apple has intentionally built its platforms to be the best in class for running AI on device, and the Unified Memory architecture Apple has created in Apple Silicon scales really well, meaning you can run ever larger LLMs on Macs. 

It’s not just Macs, either. An iPad can run up to 14 billion parameter models quite happily; a Mac Studio reaches 480 billion; and a cluster of four Mac Studios will take you all the way to 1.6 trillion parameters using off-the-shelf cables.

To put that into context, Omdia found that 57% of the AI models typically used by the enterprise come in at under 10 billion parameters, which implies that enterprises could run a huge chunk of their AI tasks on an iPad, an iPhone, and certainly on a Mac. The ability of Apple’s ecosystem to scale is precisely why most AI developers at frontier model companies already use Macs. “Organizations that build AI solutions in-house adopt Mac for AI workloads at nearly double the rate of organizations buying commercial solutions,” the report explained.

The takeaway

Apple is emerging as an important component of an overall ecosystem for applied AI in the enterprise — or anywhere else — challenging frontier models with a scalable, controllable, economical, and secure approach to deployed AI that delivers most of the bang expected for the enterprise buck. 

While Apple paid for the report, that doesn’t necessarily invalidate its conclusions, which are not myopic around the Apple platform. Apple does not replace everything else, it just becomes one of the pillars to build success with AI. Companies can use other AI services and solutions, but they’ll want Macs along for at least some of the ride. And as the models themselves evolve and become slimmer and more refined, the platforms that run them best will deliver the advantage business users need.

Now, we need Apple to develop tools for the management, deployment, and governance of these solutions.

Please subscribe to my daily, human-curated Apple-related news headline feed at The Core, or follow me on BlueSky, LinkedIn, or Mastodon.

Nvidia has released a free tool that will enable users to build an AI inferencing cluster from disparate PCs on the same network, accessible from a single interface.

Released as a beta, Nvidia Personal AI router (PAIR) connects devices running Windows, macOS or Linux to process AI inferencing workloads privately.

While the system is aimed primarily at home users, it could find favour with enterprises looking to put idle desktop compute capacity to use.

PAIR works with DGX Spark desktop supercomputers, PCs containing RTX GPUs, and some MacOS devices. The systems in the cluster run tasks in parallel, but PAIR does not turn them into a virtual GPU, Nvidia said.

The beta version of Nvidia PAIR is available for download now.

This article first appeared on Network World.

The destiny of Spirit Airline’s data is still undecided, months after the company sought bankruptcy protection.

AI data company Micro1 has now offered $12.5 million to acquire a trove of the company’s emails, Teams chats, operations and employee productivity data, according to a report by aviation website Simply Flying,

It said the data includes about 600 million email and chat records generated by 17,000 employees, as well as 17 million OneDrive files, 20.5 million SharePoint items, and more than 30 million recorded customer service calls. Such a large repository of information is a gold mine to any company looking to train AI models more effectively. The report says that this data includes sensitive, decades-old employee and workplace records.

But Micro1’s offer comes weeks after Google acquired the data at auction, with its $10 million bid beating the $7.5 million offered by another AI training company, Mercor.

The airline’s former employees objected to the sale, and last week their unions took legal action to block the it.

Nelson, international president of the Association of Flight Attendants-CWA, which continues to represent more than 5,500 of Spirit’s flight attendants, told Forbes that former flight attendants were unhappy about Spirit attempting to cash in on sensitive data when they still have not been paid their accrued vacation time, sick leave, and outstanding compensation.

However, this type of personal data is extremely valuable to the likes of Google. It means that AI models can be trained in more realistic scenarios. One option that is being explored is whether the data can be anonymized, which may offer a way forward to keep both sides happy.

This article first appeared on CSO.

Adobe’s search for a new CEO is over: It has promoted Anil Chakravarthy, president of its customer experience orchestration business, to the top role.

It has taken six months to find a successor to outgoing CEO Shantanu Narayen, who announced in March that he would step back from the CEO role, remaining with the company as chairman.

There had been much speculation that he would be replaced by Adobe’s president of creativity and productivity, David Wadhwani, who was responsible for the digital media business segment that generates around three-quarters of the company’s revenue. Having missed out on the top job, however, he has chosen to leave Adobe, announcing his departure on LinkedIn.

Chakravarthy was responsible for the development of several products, including Adobe CX Enterprise, GenStudio and Brand Visibility, as well as the introduction of CX Enterprise Coworker.

He takes over at a shaky time for Adobe: Its stock price has tumbled dramatically in the past few years as the rise of AI has meant workers can lay out content and manipulate photos without needing Adobe products. Chakravarthy, who will initially work in tandem with Narayen, will have his work cut out for arresting the current decline.

So, you’ve got email, you say? Lots of it? More than you can possibly manage without losing the few metaphorical marbles still sloshing around in that soggy ol’ brain of yours?

I hear ya. In fact, I think we all can relate (even those of us whose brains are, erm, slightly less soggy). And I’m here to tell you: It doesn’t have to be so difficult.

Gmail has a variety of built-in tools for making your messages more manageable. Some of ’em are a little bit different from what you might be accustomed to using in more traditional email clients (here’s lookin’ at you, Outlook) — but if you take the time to figure out how they work, you might just be surprised at how effective they can be.

There’s no better example than Gmail’s label system. It’s a strange concept to wrap your head around at first, especially if you’re used to the more typical folder-based method of inbox organization, but here’s a little secret: Labels actually are folders, in a sense. That, however, is just one small part of their inbox-organizing power.

Think through these nine label-centric possibilities and get ready to see Gmail’s labels in a whole new light.

1. Use Gmail labels like super-folders for categorizing your email

First, the most basic Gmail label mindset to master: You can think of a label like a folder — but with an important twist: Instead of a message being placed into a label, the label is placed onto the message.

That subtle-seeming distinction is actually quite significant. What it means is that a message doesn’t have to be associated with only one label, as is typically the case with folders; rather, you can apply as many labels as you want to any message, and each one ends up acting like a sticker — a label, one might even say! — that sits atop the email along with any other labels you’ve applied.

Any labels associated with an email will show up both in your inbox and when viewing the message in full.

JR Raphael / Foundry

So, for instance, if you keep tabs on stats for your company’s website, you might label all incoming emails from Google Analytics as “Web Reports.” But maybe you also have your own personal website for which you receive Analytics updates. You could label the reports from your personal website as “Web Reports,” too, and then add a second label of either “Work” or “Personal” onto every message to create a distinction between the two types.

On the Gmail desktop website, the easiest way to create a label is to click the label icon in the toolbar at the top of the screen when you’re viewing a message or when you’ve selected something from a message list — then start typing whatever name you want to use for the label. Once you’re done, simply hit Enter, and Gmail will create the label for you and apply it to the message.

Once you create a new label within Gmail, it’ll always show up as an option in the future.

JR Raphael / Foundry

The next time you click the label command, you’ll see your newly created label as an option.

You can also perform the same action by using the keyboard shortcut L while an email is open or selected, provided you’ve enabled Gmail’s keyboard shortcuts system.

On the mobile front, you can create a new label within the Gmail app for Android or iOS by tapping the three-line menu icon in the upper-left corner of the screen and then looking for the “Create label” option before the list of existing labels (on Android) or the “Create new” option directly after that list. You can apply an existing label onto a message, meanwhile, by looking for the “Label” (on Android) or “Label as” (on iOS) option in the main three-dot menu anytime an email is open or selected.

2. Save yourself a step and label while archiving

Here’s a handy command to remember: If you want to add a label to a message and simultaneously dismiss the message from your inbox, use Gmail’s “Move to” option.

On the desktop website, that’s the icon showing an arrow inside a folder, directly to the left of the regular label icon. On both Android and iOS, you can find the same command within the three-dot menu icon while any email is open or selected. You’ll see a list of options that include your labels; just select the label you want to assign to the message.

Whatever platform you’re using, that’ll handle all the steps for you in one fell swoop. And if you’re a fan of shortcuts like this, by the way, I’ve got a whole story dedicated to time-saving Gmail tips just like ’em.

3. Apply labels while you’re composing an email

Gmail labels aren’t only for incoming messages; you can also proactively apply them to a new message you’re composing and then know that that email and any replies associated with it will remain properly organized and present in all the right places.

On the Gmail desktop website, all you’ve gotta do is hit the three-dot menu icon in the service’s compose window and look for the “Label” option in the list that appears — then create a new label right then and there or select any existing label to apply it.

You can proactively apply a label while composing a new message.

JR Raphael / Foundry

The mid-composing label-adding option isn’t present in the Gmail mobile apps, unfortunately, but you can accomplish the same thing in a roundabout way by heading into your Sent messages (from the app’s main three-line menu) and then selecting or opening an email to add a label onto it right after you’ve sent it.

4. Organize your label lists

All right, so you’ve got your messages labeled — now what? Well, first, you can always browse through your labels to find something you need. Gmail keeps your list of labels (in alphabetical order) in its left sidebar, on the desktop website. You can collapse or expand that sidebar by clicking the three-line menu icon in the upper-left corner of the screen, and you can click on any label in the list to see all of the messages associated with it.

By default, though, that list is probably quite the unruly mess. So let’s get it organized:

• First, if you see a line in the list of labels that says “More,” with a downward-facing arrow alongside it, click it to expand your full list of existing labels.
• Now, click the three-dot menu icon next to each one of your labels. (The three-dot icon appears only when you hover over a label.) Assign each label a color, if you’d like; that’ll make it more visually distinctive and give it more or less emphasis (depending on the color you choose) when it appears in your inbox.
• In that same three-dot menu, look at the “In label list” section and consider if you want the label to always appear in your label list, to remain permanently hidden beneath that “More” divider, or to appear in the list only when unread messages are present within it. The less unnecessary clutter you have visible, the easier it’ll be to find the labels you actually use often.
• Next, look at the “In message list” section and think about whether you want the label to show up as an option when you’re adding a label onto a message. If it isn’t a label you ever actively apply to messages, you might consider hiding it to reduce clutter and increase your efficiency in that area.

On Android or iOS, you can find your labels and browse through any of ’em by looking inside the Gmail app’s main three-line menu. On Android, a third-party app called eLabels can also let you make more advanced label modifications from your phone or tablet — but in general, those sorts of adjustments are most easily managed from the Gmail desktop website.

5. Combine related labels into groups

While we’re getting your labels organized, take a moment to mull over whether any of your labels would make more sense as sublabels within a broader category. Maybe, for instance, you’d have one label called “Expense Reports” and then sublabels within it for “Staff,” “Freelancers,” and “Travel.”

This is an adjustment you’ll absolutely want to make on the Gmail desktop website, where the full selection of label customization commands is readily available. You can find the option to create a sublabel by clicking the three-dot menu icon next to any label name within the site and then selecting “Add sublabel” from the list that appears.

Grouped sublabels are a great way to keep your Gmail labels list tidy.

JR Raphael / Foundry

Just note: Somewhat confusingly, sublabels and parent labels are treated as separate but related entities — so if you were to add a message to “Staff” in our previous example, its label would show up as being “Expense Reports/Staff.” The message wouldn’t, however, show up when you click “Expense Reports” in the left-of-screen label list; it’d appear only when you clicked “Staff” in that same section. If you wanted it to appear within the results for both the parent label and the sublabel, you’d have to add both of them onto the message.

6. Use labels for smarter searches

Gmail’s labels aren’t just for browsing; you can also use them as variables to narrow down a search and make it faster to find a message you need. To include a label as a variable in a search, just click or tap the Gmail search box at the top of the website or mobile app, type label:work (using the name of your actual label name in place of “work,” of course), and then type in any other term or variable you want.

If you were trying to find a travel-related expense report from Tim Cook’s trip to Chicago, for instance, you could type label:travel from:[email protected] chicago into the search box. Including the label would narrow down your search considerably — skipping over messages from Tim Cook that might’ve just mentioned Chicago in some other context — and let you get to what you need more efficiently.

You can even narrow down your search further and search for messages that contain multiple specific labels — like label:travel label:work, which would show you a list of all messages that have both of those labels in place.

And on the mobile front, you can make your life easier yet by tapping the “Labels” dropdown within the app’s search area. That’ll let you simply select from a list of available options and save yourself the trouble of manually typing anything in.

In the Gmail mobile apps, you can select from a list of existing labels instead of typing in a search.

JR Raphael / Foundry

Gmail actually treats many of its own built-in message designations as labels, too — so in addition to searching for your own custom labels, you can search for (and in some cases also find within the mobile app dropdown menu) things like:

• label:inbox
• label:unread
• label:promotions
• label:social
• label:starred
• label:muted

Beyond that, you’d never know it, but Google also supports a hidden series of advanced Gmail labels for automatically identifying emails related to areas like travel, purchases, and finance.

And finally, any recent past searches you’ve conducted will show up as suggestions whenever you tap or click in the search box, which makes it especially convenient to revisit any frequently accessed label-related explorations in the future.

7. Let Gmail apply labels for you

Here’s where things really start to get interesting: In addition to manually applying labels to messages as you see fit, you can train Gmail to automatically categorize and process messages for you.

The first step is to think about what factors would make an incoming message fit into a particular label — whether it’s the sender’s email address or domain name, a specific word or phrase in the subject line, or maybe even the address to which the message was sent.

In my own inbox, for example, Gmail applies a red “VIP” label to emails that arrive from editors or other contacts whose messages tend to be timely and important. And it applies a green “Invoices” label to messages that come in from a specific address and with certain subject lines that my accounting software uses for carbon copies of new invoices.

In some cases, like with the “VIP” label, the messages remain in my primary inbox. But in others, they end up moving to less attention-demanding places — like the invoice copies, which skip over my inbox entirely (since I’m keeping them mostly as records and don’t need to be bothered dealing with them every time they come in).

Once you decide what factors should apply to what label — and what else should happen once the label is applied — a Gmail filter is the key to making the magic happen. I outlined the exact steps to setting one up in my separate Gmail filters guide. Go have at it!

8. Let your labels limit your notifications

Last but not least, instead of getting distracting alerts for every new message that lands in your inbox — or going to the opposite extreme and not getting any email notifications whatsoever — use your Gmail labels as a way to control which messages alert you.

You can pick any label that seems notification-worthy and then limit your notifications only to messages with that label attached. You can even set up label-driven notifications to function on your desktop computer and on your mobile device. (Provided you use Android, that is. Sorry, iPhone folks, but Apple’s garden doesn’t allow this level of customization to fly.)

That’s exactly what I do with my aforementioned “VIP” label, in fact. And it’s actually quite easy to set up. I put together a step-by-step guide here.

And with that, my inbox-dwelling comrade, your label education is complete. Go ahead: Label yourself organized. You’ve earned it.

This article was originally published in November 2019 and most recently updated in September 2026.

OpenAI launched GPT-6 Astra on Thursday, disclosing that the new flagship model has crossed the “Critical” threshold for cybersecurity risk under its Preparedness Framework, a classification the company said triggers additional deployment restrictions.

“GPT‑6 Astra is rolling out today to a limited set of organizations and over the coming days will become available to all ChatGPT Plus, Pro, Business, and Enterprise users, as well as through the OpenAI API and AWS,” OpenAI said in a statement.

Enterprise administrators must manually enable Astra for their workspace, since access is off by default at launch, according to the company.

Developers can access Astra in the API as gpt-6-astra or through Amazon Bedrock, OpenAI said, priced at $10 per million input tokens and $50 per million output tokens. Pro, Business, and Enterprise users also get a variant called Astra Pro, and the company said Astra supports Zero Data Retention for eligible API customers.

Company claims perfect score on exploit benchmark

OpenAI said it tested Astra without production safeguards on ExploitBench, and that the model scored 100%, up from 78.5% for predecessor GPT-5.6 Sol. On ExploitGym, a broader exploit-development benchmark, the company said Astra reached a 42.4% success rate against 30.3% for Sol, while using fewer output tokens.

“Its ability to identify and develop zero-day exploits can help defenders find and patch weaknesses, but it also creates a need for stronger safeguards,” OpenAI said in the blog post.

OpenAI also tested Astra on vulnerabilities disclosed in the three months before launch, to check whether it could find flaws on its own rather than recalling old exploits from training data. The model found two new zero-day vulnerabilities during that test, OpenAI said, and it is now disclosing both to the software makers involved.

Sanchit Vir Gogia, chief analyst at Greyhound Research, said the Critical label is a disclosure event rather than a capability event.

“Astra’s capability did not change between 10 August, when OpenAI said Critical capability could not be ruled out, and September 1, when it said the threshold was met,” Gogia noted. “The testing changed. The model did not.”

That inverts the obvious enterprise response, he said.

“Astra is now the only frontier model whose cyber capability an enterprise actually knows, because it is the only one measured against a published threshold, while every unlabelled model already sitting behind enterprise credentials has never been measured that way and will not be until its vendor chooses to measure it,” Gogia pointed out. “Those models are not safer.”

OpenAI said the public version of Astra will refuse advanced offensive tasks such as generating proof-of-concept exploits, though it plans to loosen those restrictions for vetted defenders through a program called OpenAI Daybreak in the coming weeks.

The launch follows OpenAI’s rollout of GPT-5.6 Sol, which the company said scored 73.5% on ExploitBench at launch, and comes months after Anthropic’s Fable and Mythos models were briefly pulled from export markets over similar concerns.

Governance shifts from the model to the harness around it

Gogia said the bigger shift is that reasoning now translates into state change, since a wrong chatbot answer is an information problem while a wrong agent action inside a customer-record system is an operating event.

“The governance unit therefore moves off the model,” he said, arguing the relevant question is no longer which model is approved, but how much damage a given identity can do before a control intervenes.

Amit Kumar Jena, head of AI development at Kanerika said the visibility problem is concrete: when an agent acts through a user interface, systems of record log the action as a person, so an agent that updates 400 ERP rows shows up as a service account making 400 updates, with no record of which instruction or model version produced them.

“You lose granularity inside the exact system a regulator or auditor will ask to see,” Jena added.

OpenAI said it built a new evaluation, informed by an incident involving Hugging Face, to test whether a model given an impossible task would exceed its authorized scope.

“Compared to GPT‑5.6 Sol, which without production safeguards went beyond the authorized target 48% of the time, GPT‑6 Astra did this in 0% of cases,” the statement added.

Gogia said the more uncomfortable finding is that Astra behaves better and watches worse: OpenAI reports decreased chain-of-thought monitorability against Sol, with Astra less likely to reveal incriminating reasoning, and its monitoring covers OpenAI’s own external deployment but nothing published extends that telemetry to customers. “OpenAI being able to monitor Astra does not mean an enterprise can audit Astra,” Gogia said.

The article originally appeared on CSO.

Enterprises are facing a disturbing new question in the age of AI: What happens when agentic assistants go dark?

This became a very real scenario on Thursday, as OpenAI’s ChatGPT, Anthropic’s Claude, and SpaceXAI’s Grok near-simultaneously, and somewhat mysteriously, experienced significant, prolonged outages.

Beginning in the morning, Eastern time, several ChatGPT models went down over a roughly two hour period, Claude models over a four-hour span, and Grok models for a near three-and-a-half hour duration. All three companies acknowledged the “elevated” issues and applied fixes.

As users grumbled in forums and IT teams scrambled to get them back online, the incident revealed how hastily some organizations have adopted generative AI workflows without considering the potential, and inevitable, impact of widespread outages.

AI agents are increasingly taking over automated and wider-scale workflows, and enterprises could find themselves “uncomfortably exposed” when AI hits the brakes, said technology analyst and journalist Carmi Levy. The situation should “serve as a wakeup call to IT leaders who have largely ignored what it’ll cost them if these increasingly critical platforms suddenly go dark. The risk is no longer hypothetical.”

Hours-long outages impact core services

ChatGPT went down on the same day as OpenAI’s anticipated launch of GPT-6 Astra, the new frontier model that the company says approximates artificial general intelligence (AGI) and gets nearer to its goal of creating autonomous systems that outperform humans.

The OpenAI outage occurred around 11 a.m. ET on Thursday and impacted a slew of services, including search, file uploads, agents, GPTs, voice mode, image generation, ChatGPT work, Compliance API, Deep Research, ChatGPT Atlas, and other connectors and apps. In some cases, users were prevented from logging in, conversations failed to load, and the interface returned errors when attempting to send messages. OpenAI’s Codex services, including web, API, command line interface (CLI), and VS code extension, were also impacted.

OpenAI fixed the issue by 12:55 p.m. ET, and advised Codex remote control users to re-pair their mobile devices.

Claude began to go dark around 7:37 a.m. ET, with Anthropic acknowledging an “exhaustive list” of impacted models with elevated errors over the next few hours: Mythos and Fable 5.1 and 5, Sonnet 5, and Opus 5, 4.8, and 4.6.

The issue was resolved by 11:27 a.m. ET. The incident followed a roughly 27-minute outage just the day before, also due to elevated errors on requests in Sonnet 5.

Grok, meanwhile, began experiencing issues around 9:30 a.m. ET. Grok Web, Build, API, Office/Workspace plugins, Android, and X were all impacted. The services returned to “healthy” traffic at 1:08 p.m. ET.

“It’s a curious scenario for multiple different providers to experience outages at the same time,” noted Brian Jackson, a principal research director at Info-Tech Research Group. It could be related to a common infrastructure such as a content delivery network (CDN) layer, domain name system (DNS), or shared cloud infrastructure, he theorized.

A case for outage planning

Just a few months ago, the extent of AI use within the typical enterprise was limited to employees using chatbots to get answers to basic questions or to draft simple email messages, Levy noted. Large-scale AI platform outages, when they occurred, had relatively little impact on overall organizational productivity. “But things are changing, and quickly,” he said.

Organizations must now have a better understanding of the impact agentic AI has on day-to-day workflows, and the degree to which they disrupt employees’ ability to complete complex tasks once they’ve handed the reins over to automated, cloud-based tools, Levy noted.

In incidents like Thursday’s, employees may fall back on traditional manual workflows, such as updating spreadsheets or pulling reports together the old-fashioned way. But they might also realize that, after relying on AI agents to do so much work on their behalf, they’ve become too dependent on automation, and their “cognitive skills may not be as sharp as they once were,” Levy said.

The growing prevalence of agentic AI should prompt organizations to revisit their disaster recovery and business continuity plans and assess the productivity impact of potential service outages, he said. While cloud-based productivity platforms like Google Workplace and Microsoft 365 offer limited degrees of “offline mode” functionality using locally-stored data, and documents can be synchronized to hard drives in Dropbox or Google Docs for Desktop, agentic AI platforms offer up fewer offline workarounds, at least in their current form.

Organizations should document workflows in greater detail and scenario-plan what near-term recovery might look like in the event of an extended AI platform outage, Levy said. They also need better training to ensure employees maintain their manual skills over time and are equipped to press them into service in the event of a service outage, because the more enterprises lean on agents to complete critical tasks, “and pull humans out of the loop in the interest of productivity,” the less able employees will be to step back in during inevitable service interruptions, he pointed out.

“It is entirely possible for otherwise well-meaning organizations to be over-reliant on AI automation,” Levy said. “Too many organizations are about to learn some hard lessons about not having a backup plan in place.”

Info-Tech’s Jackson also recommends a modular architecture for LLMs; enterprises should view the model as a “commodity that can be hot-swapped with an alternative.” That might be another cloud service provider (which hopefully isn’t experiencing a concurrent outage) or a self-hosted option like an open-weights model.

“In a scenario like this, when your first choice provider might not be available, you have a fallback that can supply that same intelligence layer, even if it’s only a stopgap solution,” said Jackson.

Microsoft’s text suggestion feature will now be turned off by default in both Word and Outlook, reports The Register. The text suggestions attempt to predict which words or phrases the user intends to type in advance.

According to Microsoft, some users appreciate the feature, while others find the suggestions distracting. With it disabled by default, interested users can now choose for themselves whether they want to turn it on manually.

It is unclear whether the change will disable text suggestions for existing users who already have the feature enabled, or if it applies only to new installations and profiles.

The change applies to Word for Windows, the web, iOS, and Android, and to classic Outlook for Windows and Outlook for Mac. Rollout timing will be communicated through the Microsoft 365 admin center and release notes, Microsoft said.

This article originally appeared on Computer Sweden.

Related:

• Word for Microsoft 365 cheat sheet
• Outlook for Microsoft 365 cheat sheet
• More Microsoft cheat sheets

A serious vulnerability was recently discovered in Exchange Server 2016, Exchange Server 2016, and Exchange Server Subscription Edition (SE). The vulnerability is designated CVE-2026-62911 and can be exploited by hackers to gain full access to affected systems, according to Bleeping Computer.

Microsoft has released security patches to address the vulnerability as part of its August 2026 Patch Tuesday release, but there are still 21,899 unpatched servers at risk, according to The Shadowserver Foundation. The highest concentrations of vulnerable servers are in the US and Germany, the security group said.

The Netherlands National Cyber Security Centre and other agencies have urged admins to install the latest patches as soon as possible.

This article originally appeared on Computer Sweden.

Related:

• Exchange CU1 delayed further as Microsoft races to verify AI-found flaws
• Microsoft Exchange Server on prem gets a little harder to use
• Is it time to move to hosted Exchange? Considerations for IT

I’ve been writing about Apple and how it benefits the enterprise for so long it’s easy to forget that the company doesn’t make enough noise about its accomplishments. 

Sure, it has pursued and won the argument about Total Cost of Ownership, proving that while initial costs might be higher, dollar for dollar Macs are a far more cost-efficient platform, particularly when it comes to product reliability and tech support. 

Apple already offers so much

Apple has also introduced extensive tools and APIs to help manage enterprise Macs, spawning a wide ecosystem of providers — most visibly, Jamf. Certain Apple products are already becoming deeply ingrained in some key professions; just think about Vision Pro and what it offers in medical care or prototype design, or the tens of thousands of Macs being carried around by AI developers across all the big name firms.

Some of what Apple delivers is already of major benefit to enterprise users. MLX for example, is being actively used to support real-life AI implementations, including by active AI-based businesses such as Yembo. And simplicity and ease-of-use isn’t just an advantage to consumer users, it boosts productivity in enterprise, too.

Almost 10 years ago, the focus was all about mobile enterprise, something that hasn’t gone away. Indeed, it is arguable that with Siri AI and third-party AI partners, Apple’s mobile enterprise story has become even more compelling. All the same, even then it was crystal clear that Macs had a big part to play in the future of enterprise tech. The iPhone accounted for 72% of all enterprise smartphone activations back then, while Jamf data has consistently shown us the steady forward momentum Mac has in business. 

So why hide the light?

As former Apple Enterprise Marketing Manager Todd Dailey points out, Apple does have some people it trusts to tell its business and enterprise stories. But it isn’t investing very much in making sure they have stories to tell. For example, he notes just one enterprise-related story on the Apple website. Published in Chinese only, that story is about Haidilao, which is seeing serious TCO and energy consumption benefits by running its back-end operations on Mac minis.

He also points to a near-mythical Apple-in-business event the company ran in Cupertino last June. That event generated almost no coverage anywhere, because no one was there to report on it. And yet it gathered leaders from Disney, Ford, Anthropic, Perplexity, Christie’s and presumably elsewhere to talk about how they use Macs in business, many with a focus on enterprise AI. I’ve spent time trying to track down additional information from that event, but there is not much out there. And while I’m willing to accept that some of those who took part needed business confidentiality, it was a semi-public event, so it seems unlikely significant secrets would have slipped out. 

It feels like a lost opportunity. Imagine the follow-up if Apple had paid a media team to attend the event to churn out stories, develop case studies, shoot video, and make viral influencer tik-toks. Is it that Apple doesn’t believe in its own enterprise offer? I don’t think so. It has a small army of business experts available to customers in stores, and they wouldn’t be there if it didn’t see a need for them.

Overcome myopia

Dailey thinks it’s a disconnect generated by Apple’s traditional focus on consumer markets. He’s probably correct, but it is frustrating; the real value Apple offers enterprise IT has been crystal clear for years – certainly since before then-Apple CFO Luca Maestri declared that Apple had set a new enterprise revenue record back in 2017. “Corporate buyers reported a 96% satisfaction rate and a purchase intent of 68% for the June quarter,” said Maestri at that time.

With that kind of momentum across such an extensive length of time, the company has without doubt built strong foundations of enterprise success. But these emerge most frequently as short footnotes in CFO statements during fiscal calls, rather than being shouted from the rooftops.

Here are some details

Highlights announced during those calls since 2024 include:

• Nvidia launched a Mac-as-choice program with more than 10,000 Macs deployed worldwide.
• UC-San Diego Health became the first hospital in the world to test Vision Pro spatial-computing apps in clinical surgical trials.
• BMW Group deployed tens of thousands of iPhones, including to factory employees.
• Capital One expanded its “Mac Choice” program with thousands more MacBook Airs.
• Crédit Agricole (France’s leading retail bank) turned to on-device AI on the MacBook Pro to cut regulatory-workflow processing time by more than 80%.
• Perplexity selected the Mac as its preferred platform for building enterprise-grade AI agents.

Apple knows these wins exist, and recently launched Apple Business, the all-in-one platform that combines hardware, software, and enterprise services to manage large-scale deployments. This showed the company knows what’s going on.

Mac does AI

Follow the money and it feels as if the next stage of the Apple-in-the-enterprise journey will at least in part be built around AI; Apple has major advantages it should celebrate with the sector. It offers the best systems for on-device AI, use and development. MLX is unique, ahead of its time, and worth leaning into. Its commitment to privacy is becoming increasingly and recognizably important to enterprise professionals. Even its battles to protect encryption are fundamental to business success.

Lots of its customers, not just Perplexity or Crédit Agricole, already see the advantages.

The successes it already has should be celebrated on the company’s own enterprise websites, and the company should recognize and invest in those stories and share them. Dailey points out that developers at Nvidia, Anthropic, OpenAI, and most enterprise AI shops all already use MacBook Pros for portable AI, suggesting a campaign around “Mac Does AI” should be in place. I see his point. I hope Apple does.

Now please subscribe to my daily, human-curated Apple-related news headline feed at The Core, and follow me on BlueSky, LinkedIn, or Mastodon.

With the launch of a new integration this week, Adobe is making more than 70 Creative Cloud applications available in Slack, promising to bring generative AI (genAI) document creation closer to team conversations. 

The Adobe MCP (model context protocol) integration allows Slack users to access, edit, and create new documents from Adobe creative and productivity tools without switching apps. This includes Photoshop, Premiere, InDesign, Lightroom, and others.

The integration should help “improve efficiency and speed of content creation,” said Irwin Lazar, principal analyst at Metrigy. “We have consistently found in our research that companies want to use their team collaboration apps, like Slack, as work hubs.” 

To access the integration, users interact with Slackbot — Slack’s AI “teammate” — via natural language prompts. 

Adobe outlined several use cases in a blog post Wednesday. It’s possible to turn “project plans and conversations” in Slack into “PDFs, images and videos you can share with your team in Slack,” and turn speadsheet data into a “PDF or visual you can share with your team or client.”

Other uses include the ability to edit images such as headshots and event photos in bulk — with controls available to adjust lighting and tone, crop and resize images, and more — as well as to search for Creative Cloud assets from Slack.

The editing interface is more limited than in the full Adobe apps, which Adobe said users can switch to when they want “greater creative control” and “pixel-level precision.”

The Adobe integration is available to Slack Business+ and Enterprise+ customers. Adobe said that most Adobe apps are “currently available for free to users,” though it didn’t confirm whether this is a temporary or long-term pricing structure. The integration includes access to some generative AI features, with daily usage limits, though others, such as generative video, require paid Adobe accounts.  

Overall, the move is part of a wider effort by Slack to make AI tools available where work conversations occur. That was the idea behind the recent Slack Code launch, for instance; it’s aimed at aiding interaction with AI coding assistants for multiple coworkers around one-off software development tasks such as bug-fixes and feature updates.  

For Adobe, it’s a continuation of efforts to make its software available in other platforms. The company recently announced ChatGPT integration that lets users create and edit content from the AI assistant. 

The Adobe integration with Slack allows conversations and work to “move closer together,” said Joe Cicman, principal analyst at Forrester. “Instead of Slack being where requests are discussed and Adobe being where changes are made, the conversation itself can increasingly become part of the creative workflow,” he said.

“More broadly, we’re seeing enterprise software move toward bringing capabilities to the worker rather than forcing the worker to move between applications,” he said.

Cicman expects that the ability to interact with Adobe tools in Slack will appeal first to “hands-on marketers and content creators who want to prototype ideas quickly or generate variants of existing assets.

“More broadly, it appeals to people who have more ideas than they have expertise in Adobe’s professional tools,” he said. “They know what they want to create, but not necessarily how to navigate Photoshop, Premiere, or Illustrator. Conversational interfaces let them express intent instead of mastering the UI.”

There are other considerations about an integration that enables users to generate content with greater ease. It could, for example, increase the production of low-effort, AI-created documents within organizations. 

According to Adobe, the goal of the integration isn’t only to help workers create more content, but to “express their ideas, contribute to the creative process and bring those ideas to life faster.

“By making Adobe’s pro-grade tools accessible through conversation, people across teams can more easily explore ideas, build on each other’s thinking and collaborate on how to make the work better,” said Deepti Pradeep, senior director for Agentic Product at Adobe. “At the same time, making creation easier only makes human judgment more important. Adobe’s tools can accelerate the creative process, but people still bring the ingenuity, taste and point of view that distinguish great work. 

“For brands, that human perspective is ultimately what turns content into something that stands out and creates a meaningful connection with an audience.”

Easier content generation could require companies to more closely manage an acceleration in AI-generated outputs,” said Cicman. “What’s happening across marketing and creative teams is that prototyping is becoming dramatically faster,” he said. 

“A marketer with a concept can describe it in Slack, AI can generate several potential executions, and suddenly the organization has many more ideas moving through the pipeline. The challenge is that faster creation creates a pile-up of work that needs to be reviewed, refined, and approved.” 

A natural outcome from reducing the cost of prototyping, said Cicman, is the emergence of a new operating model for teams. 

“Historically, designers spent a significant amount of time creating initial concepts from scratch. Increasingly, AI can generate those first drafts,” he said. As a result, a designer’s role “shifts toward refinement, curation, editing, and applying taste.

“In that model, the marketer contributes the concept, AI produces the prototypes, and the designer elevates the work from ‘technically generated’ to ‘creatively excellent,’” he said. “The organizations that benefit most won’t be the ones that generate the most content. They’ll be the ones that combine abundant AI-generated prototypes with strong human judgment and creative taste.”

Former Apple chip supplier Intel has faced a lot of challenges since Apple ceased to be a client. It struggled to win back some of Apple’s processor manufacturing business, but seems to face one fundamental challenge — trust.

That’s the very briefest gist of an extensive and outstanding report today from Culpium, which gives us an insider look at the relationship between the firms. It’s particularly relevant, given that the Trump Administration has announced that Apple agreed to work with Intel in the future. (TSMC is Apple’s current supplier and that relationship seems very solid.)

The root of the challenge

The problem with Intel, according to the report, is that a lack of trust drove Apple to seek an alternative processor supplier in the first place. The report discusses some of those challenges, including Intel’s failure to keep the pace with Mac chips, which is why the M1 chip astonished the industry, because it gave Apple the performance it had always sought.

How we got here

The root of that release was mobile, of course. Apple’s A-series chips were built for mobile and eventually transcended to become M-series Mac chips. Intel turned Steve Jobs down when he asked it to make mobile processors, pushing Apple to TSMC. The history is more complex than that, of course — Apple also acquired chip development expertise from PA Semi, for instance — and established a foundry deal with TSMC that meant the company produced Apple-designed chips. The work Apple did on mobile processors eventually enabled the move to Apple Silicon.

That’s the history. Fast forward to today and an ailing Intel needed new client wins even as the Trump Administration sought to support onshore processor production; it’s a strategic need for the US. The problem is, as Culpium terms it, that Intel isn’t willing to build manufacturing capacity until it wins the supply contract. And Apple needs a much firmer promise than that to supply its huge mass market. It needs capacity before it can provide trust.

Telling tales too soon?

During the most recent negotiations concerning Intel’s capacity to deliver chips in quantity, the report tells us Intel somehow managed to challenge that trust by telling third-party partners that it had secured Apple as a client. Anyone who knows Apple will recognize that the company doesn’t like its business discussed that way. “A sure sign that the two companies may be on the path to some serious cultural clashes,” wrote Culpium.

TSMC, meanwhile, has a corporate commitment to trust, a promise recognized across the industry — even by Nvidia founder Jensen Huang. That trust extends into tomorrow, which is why TSMC invests vast sums in developing process technology it believes customers will need tomorrow, rather than waiting for them to order up capacity today. That’s also why TSMC today offers some of the world’s most advanced interconnects, enabling powerful tech that likely includes Apple’s M5 Ultra. But just as important as technology and manufacturing capacity is a trusted relationship between companies. 

That’s not to say capacity doesn’t matter. It does. And as Apple continues to sell its products in ever greater quantities, it knows it must source additional supply; that’s why it speaks to Intel, Samsung, or even CXMT about the components its products require. But ultimately, a company that historically has felt let down by numerous key competitors — think Google, Samsung, and Android or, more recently, elements of OpenAI — will place a great deal of value in discretion and corporate trust.

After all, Apple has learned that in the ultra-competitive market in which it exists, what some call paranoia will in the end be seen as business sense.

You can follow me on social media! Join me on BlueSky,  LinkedIn, Mastodon and subscribe to The Core.

When it comes to productivity these days, we’re got two competing concepts to consider:

1. Automate everything with AI and let an unreliable, error-prone bot do work on your behalf (thus likely spending even more time reviewing — and correcting — its mistakes)
2. Find intelligent systems for increasing your own efficiency while maintaining complete control over what you’re doing

The first path clearly gets the most attention as of late. (And, to be fair, there are ways that AI can genuinely be useful, especially when it’s used as a tool for specific limited purposes as opposed to an omnipresent answer for everything.) But arming yourself with good old-fashioned efficiency-enhancers is an increasingly underrated and overlooked option for giving yourself a powerful productivity advantage — in a way that actually makes a difference, without any awkward compromises.

That’s particularly true in the domain of the desktop browser, where so many of us do so much of our work in 2026. The browser in many ways has become our de facto workday operating system — no matter what kind of computer we’re using at any given moment — and yet, it’s nowhere near the level of an actual operating system when it comes to the shortcuts and work management tools it offers.

At this point, the company behind the world’s most widely used browser is seemingly more interested in finding new places to cram Gemini into our lives than developing genuinely useful systems for more efficient browser work. But a third-party developer we’ve talked about before in these quarters is stepping up to fill that void and give Chrome the keyboard-centric kick-in-the-keister it needs to become an optimal productivity backdrop — with our own fleshly fingers guiding the way.

It’s a brand new browser extension that’ll work in any Chromium-based browser and completely change how you think about getting around your tabs and getting stuff accomplished. I’ve had the opportunity to test it out and live with it for several weeks now. Let me show you how it works and what it could do for you.

[Get level-headed knowledge in your inbox with my free Android Intelligence newsletter. One smart new thing to try every Friday!]

Meet your new Chrome command center

My fellow keyboard warrior, allow me to introduce you to a little something called Tab-da.

Tab-da is an extension for Chrome — or, again, any other browser based on the same underlying code, including Vivaldi, Edge, Brave, and Opera — that adds all sorts of keyboard-oriented productivity-minded power-ups into your browser environment.

There’s so much to it, in fact, that the biggest challenge is wrapping your head around all of the possibilities it presents and developing the muscle memory to get yourself in the habit of using actually ’em.

The best way to get going is to start small — with a focus on a few core Tab-da features:

1. The tab switcher

Tab-da’s centerpiece, so to speak, is the simple yet robust OS-like tab switcher it adds into the browser arena. Just like you how you hit Alt-Tab to move among programs in Windows, you hit Ctrl, Shift, and the left or right arrow key with Tab-da in your browser — and, well, ta-da:

Tab-da’s on-demand tab switcher in action — a faster, easier way to get where you want to go in your browser.

JR Raphael, Foundry

You can zip between all of your open browser tabs as if they were apps or processes on your computer — ’cause for all intents and purposes when working these days, they really are. And now, at long last, they can act like it.

2. The tab command bar

Perhaps the most powerful part of Tab-da is the all-purpose launcher it adds into your browser for almost any action imaginable. No matter what else you’re doing at any given moment, with Tab-da installed, you can hit Ctrl, Shift, and the spacebar together to summon a command bar that’s reminiscent of the Ctrl-K (or ⌘-K) option in many desktop apps and just overflowing with productivity potential.

Tab-da’s command bar is the keyboard shortcut command center Chrome’s been missing.

JR Raphael, Foundry

From the prompt, you can start typing the name of any website or web app you’ve got open to find and jump to it quickly — or just use the on-screen numbers to move instantly to any site you see. You can also tap into a handful of handy keyboard shortcuts to zoom in specifically to other browser-based areas:

• Alt-O will pivot the command bar to showing and letting you search through tabs open within the same browser on other devices, like your phone.
• Alt-B will allow you to browse or search through your saved bookmarks.
• Alt-H will take you to your cross-device browsing history.
• Alt-C shows you recently closed tabs.
• Alt-D empowers you to see and search through your downloads.
• And Alt-T opens an action-packed actions menu where you can move a tab, pin it, duplicate it, copy its URL, or share it all right then and there — all without ever lifting your fingers off your keyboard.

Actions in the command bar are a convenient way to do more than just moving between tabs.

JR Raphael, Foundry

And those are just the default built-in shortcuts.

3. Custom keyboard shortcuts

The part of Tab-da I might appreciate the most is the framework it gives you for creating your own custom keyboard shortcuts for browser-based actions both simple and complex.

Within that same command bar we were just going over, you can set up quick keyboard shortcuts for any sites you find yourself opening often — then get to them simply by hitting Ctrl-Alt-space and entering whatever shortcut you assign them.

Any sites or combination of sites can be kept a couple quick keystrokes away with Tab-da’s custom shortcuts system.

JR Raphael, Foundry

What’s especially cool is that you can also set up multiple sites to open together in this same way — if, for instance, your workflow involves opening a specific series of four sites together (say, Google Docs, WordPress, Trello, and Notion) to work on one particular task. Then you can always get to that work mode with a few quick keystrokes and no manual tab-by-tab opening.

Your Chrome keyboard control adventure

Tab-da has tons of other time-savers that you’ll keep discovering the longer you use it, but those three areas are the perfect place to start — and the perfect way to understand what the extension is ultimately all about and how it aims to help you.

Oh, and as for the developer? As I alluded to at the start of this story, he’s someone you may know if you’ve followed my suggestions around Android for long. His name is Bardi Golriz, and he’s the same guy behind both the Ruff Reminders app I recommended earlier this year and the Ruff Writing widget I’ve written about before.

Like Golriz’s other apps, Tab-da doesn’t contain any kind of tracking or other disconcerting privacy asterisks. All of your browsing data remains entirely on your device, with no syncing, storage, or sharing of any sort.

The extension is free for an unlimited seven-day trial — with no payment details or personal info required. After that, if you want to keep using it, it’ll run you a whopping 99 cents a month or 20 bucks for a lifetime license, which is a small price to pay for all the seconds it saves you (and, of course, a price that lets independent developers like Golriz continue doing the work they do).

Just remember: Start small. Let yourself get in the habit of using one feature and one keyboard shortcut at a time, then add one more into the mix from there as each element slowly becomes second nature.

Before long, you’ll barely remember what working in a browser was like without a tool like this in the picture — and you almost certainly won’t want to go back.

Discover oodles of useful tech tricks with my free Android Intelligence newsletter — one new thing to try every Friday, straight from me to ye.

Learning from the OpenAI-Hugging Face fiasco, as well as from recent revelations about its own model, Anthropic is revamping its security and alignment practices.

The company has established controls that flag when a model attempts to break out of a sandbox or successfully accesses the live internet, cordoned off its highest-risk test environments, and proposed a set of safety standards for its external testing partners, such as giving AI agents explicit instructions like “you should not access the internet.”

Anthropic conceded that three recent security incidents involving Claude reflect a “failure of operational security,” and also reveal issues with model reasoning capabilities and “recklessness.” Recent events “stressed that the urgency of improving our cybersecurity defenses is even higher than we previously believed,” the company noted.

Anthropic’s approach to security and alignment

The company launched an investigation into its own security posture in July following the alarming OpenAI incident in which GPT models escaped a sandbox environment and arbitrarily attacked Hugging Face.

The company subsequently disclosed three situations during cybersecurity testing in which Claude models (Opus 4.7, Mythos 5, and an internal research model) accessed computer systems they should not have been allowed to touch.

The pre-release models were intentionally running without cyber safeguards, a common practice in early testing, and were able to exploit misconfigurations in a third-party’s settings using basic hacking techniques. Flaws in their reasoning led them to believe that all the entities they accessed, including those on the live internet, were in-scope for their capture-the-flag exercise.

Anthropic maintained that its internal security posture was not a contributing factor. The exploits occurred in a third party environment where internet access was mistakenly left open, so “the models had no need to ‘hack out’ of anything, even if they had been inclined to do so.”

Still, the incidents underscored the importance of hardening the sandbox itself, the company said. Until now, builders had been “largely relying on a single layer of defense,” that is, the configuration of the environment, when they should have had several layers in place, including monitoring, setting explicit boundaries within prompts, and sealing sandboxes.

Following its discoveries, Anthropic paused its internal and external evaluations of pre-release models. Higher-risk reinforcement learning (RL) environments for pre-release models were also halted for several weeks, while some sandboxes were moved to isolated settings with more stringent security gating.

Researchers built and deployed a classifier that automatically identified a model’s attempts to “aggressively probe” or break out of a testing environment, and when it inadvertently gained access to the live internet. They also inspected previous model evaluations to identify sandbox escapes or misconfigurations.

The company said they found no examples of models breaching sandbox boundaries. And while several sandbox configurations were flagged, none resulted in a compromise of outside systems.

When it comes to alignment, however, Anthropic researchers discovered a common failure with motivated reasoning: Models believed they were still in a simulated environment when they were actually connected to the internet; thus, they did not think they were breaching the boundaries of their testing lab. The other flaw was “recklessness,” a model’s seeming willingness to take harmful actions in the wild to ensure it achieved its goals.

In response, the researchers have been resampling models, testing them in skewed settings, and filtering out environments that incentivize cheating. They have also overhauled their production RL stack, established a stricter review process, changed specifications around model rewards, and tightened criteria for human reviewers who often dismissed false positives.

Further, Anthropic has expanded its monitoring of internal employee usage, and continues to red team its stack. Some high-risk RL environments remain paused for manual review.

Researchers are still studying how, exactly, models determined whether or not they were connected to the live internet, and if they knowingly disregarded signs that they were, due to flaws in their reasoning. They have also “empirically” determined that defects in training sandboxes disproportionately contribute to misaligned behavior.

However, that is not the sole reason. “Solving alignment will involve addressing a very wide range of potential problems,” Anthropic noted, “and future alignment incidents may involve different behaviors and different causes from those we have seen so far.”

Testing best practices

Because the Claude security incidents occurred in third-party environments, and pre-release models often have weaker security safeguards, Anthropic is asking external testing companies to commit to a set of best practices.

Notably, they should explicitly set a model’s scope, and define permitted actions and what a model should and shouldn’t access. Boundaries must be stated as instructions (“you should not access the internet”) as opposed to descriptions of the environment (“you do not have internet access”).

Other suggested practices include running continuous, real-time monitoring, conducting evaluations inside a hardened sandbox with no internet access, and probing for vulnerabilities before even performing any testing, instructing the model itself to attempt to break out of its sandbox, and having it report back if it succeeds. Further, testers must confirm that evaluation challenges are actually solvable, rather than theoretical, so models don’t take actions outside their intended scope to achieve them.

“Because evaluations are sometimes conducted over hundreds or even thousands of runs, partners should ideally carry out this testing at length before beginning any evaluations,” Anthropic noted.

The company said it is developing companion best practices for those with access to Claude Mythos 5, which also runs without cyber safeguards.

Going forward, Anthropic described a “defense in depth” strategy. During alignment, a model is trained to be “helpful, honest, and harmless,” and is steered away from irreversible or contextually irrelevant actions. Models are given minimal permissions and their actions are limited, while offline monitoring notifies humans when things look wrong.

Finally, as a last resort, risky actions are blocked based on pre-determined classifiers, and humans can “pull the cord,” rework, or pause an agent when security layers fail.

Safety is just one part of it

Experts call the move a positive step, if a basic one. Best practices like better isolation and monitoring should have been in place before agents were kicked off to hack systems, noted David Shipley of Beauceron Security.

“Better late than never,” he said, adding: “All these frontier firms are benefiting from felony-humblebragging-as-marketing, but there are some solid improvements in this announcement.”

The fact that the EU Act is now in force adds another layer of context, Shipley pointed out: Europe’s regulators are digging into the safety issues posed by frontier AI. These companies have had one of the fastest growth trajectories in tech history, and, concurrently, arguably the fastest regulatory response. Ideally, regulators are taking lessons from the “social media mess” and staying on emerging tech’s case before massive harms ensue, he said.

At the same time, frontier AI companies are watching high-profile court cases like the one targeting Meta.

This adds a third layer of context: The speed at which these companies are being sued is also on an unprecedented trajectory. “So, we should also read this blog as building a paper trail for a due diligence defense for regulators and courts,” Shipley noted.

This article originally appeared on CSOonline.

Citrix on Tuesday announced that it has completed the acquisition of longtime partner Numecent, producer of technology that containerizes and manages Windows applications.

The acquisition builds on joint efforts to integrate Numecent’s management tool, Cloudpager, with Citrix Desktop-as-a-Service (DaaS) after an integration announced in April let administrators natively publish and manage the application containers through familiar Citrix workflows.

Numecent’s other product, Cloudpaging, packages Windows applications into isolated application containers independent of the underlying operating system, streaming them to Windows endpoints on demand rather than requiring them to be included in a desktop image. 

Citrix plans to further integrate the technology into its platform, while also continuing Cloudpaging and Cloudpager support for physical Windows devices.

“Enterprise customers have told us for years that application management is one of the most painful parts of running a Windows environment,” said Shawn Bass, SVP and GM of Citrix DaaS, in the announcement of the acquisition. “Numecent has solved this in a genuinely elegant way. By bringing Cloudpaging and Cloudpager into Citrix, we can make this capability native to every DaaS and physical desktop deployment so IT teams get back the time they spend wrestling with images and app conflicts.”

Analysts and consultants said the move will help enterprise IT to some extent, but will also increase vendor lock-in with Citrix while potentially exposing enterprises to data security risks.

Good for Citrix customers

Gartner VP Analyst Stuart Downes said, “overall, this is a positive for Citrix customers,” but he stressed that the promised conversions “are not 100% compatible.” 

He said, “low-level integrations into the kernel are generally not successful” because code that needs the lowest level of OS integration usually needs direct links to the hardware. Still, he estimated that applications at the low level probably account for only 2% of enterprise applications. 

For the more typical apps, Downes said that there will likely be “north of 90% compatibility. It varies. There are quite a lot of complex factors in app virtualization.” But he emphasized that Numecent offers two components: Cloudpager and Cloudpaging, and “we have yet to see how Citrix will integrate both.”

Justin Greis, CEO of consulting firm Acceligence, also sees a lot of potential savings for the enterprise.

“Large companies can have thousands of Windows applications, including legacy, custom, industry-specific, and highly specialized applications,” he said. “Many have dependencies on particular versions of Windows, libraries, configurations, or desktop images. Every major desktop refresh, Windows migration, VDI program, cloud move, acquisition, or infrastructure modernization effort can therefore create another application testing and repackaging cycle. The ability to abstract more of the application layer from the environment underneath it can remove a meaningful amount of that friction.”

Noah Kenney, principal consultant at Digital 520, added that the theoretical advantage that Citrix can now offer has great enterprise potential.

But, he argued, this likely amounts to an enterprise IT pay less now, pay more later situation.

“There are operational savings here, which is why customers will adopt it, but the bill comes due when they try to leave,” Kenney said. “This is a good acquisition for Citrix and probably bad for enterprise leverage over time. Citrix can now lose the desktop and still keep the customer. Every application moved into Cloudpager raises the cost of the next migration. Customers get the simplification now and Citrix gets the switching cost later.”

Half right

Sanchit Vir Gogia, chief analyst at Greyhound Research, said that he reads the containerization pitch as half right. “The packaging premise is valid. The cross-operating-system execution premise is not,” he said.

Gogia pointed out that Numecent Cloudpaging packages a Windows application with its dependencies and streams it to a Cloudpaging Player on a physical or virtual Windows endpoint, where it executes locally. “A Mac or Linux user reaches that application through Citrix’s remote delivery, where it still executes on Windows. That is cross-platform access, not cross-platform execution,” he said. “A Windows application does not become a Mac application merely because its pixels arrive on a Mac. The container is a packaging promise and the boundary of that promise is Windows.”

That said, he noted that there is still a lot of value in the Citrix arrangement, because Cloudpaging separates an application from a particular Windows image and carries that package across physical and virtual Windows environments, including Arm-based devices. 

“The real advance is not escaping Windows,” Gogia explained. “It is making application change less dependent on desktop change. Microsoft’s own App Assure data puts enterprise application compatibility above 99.7%, and Cloudpaging’s commercial logic lives almost entirely inside the fraction that remains. At enterprise scale, the final 1% of applications can carry far more than 1% of the business risk.”

But, he added, “Existing Numecent customers need binding answers on entitlements, migration and exit. Citrix has bought control of a useful Windows application lifecycle. Control now has to prove itself, and the proof it owes customers is less complexity, not merely more control for Citrix.”

Possible risk

However, consultant Brian Levine, executive director of FormerGov, pointed out that the nature of these new Citrix capabilities could expose users to serious security issues, including the risk of data exfiltration. 

He sees Cloudpager as “essentially a privileged switch that can push software to every Windows endpoint at once, which is precisely the kind of mass-distribution channel that produced SolarWinds and Kaseya. Bolting it onto Citrix, whose NetScaler gear has been a favorite ransomware target through repeated ‘CitrixBleed’ flaws, may leave CIOs and organizations wondering who will focus on security for the combined entity, and how will it prevent the type of attacks we’ve seen against Citrix.”

Citrix was asked to comment on these security questions, but did not do so by publication time. 

Older employees are significantly more positive about AI than their younger colleagues, according to a new survey by Glassdoor.

Among members of Generation X (those born between 1965 and 1980), nearly half of respondents had a positive view of the new AI technology. At the same time, just one-third of Generation Z (that is, those born between 1997 and 2012) share that same level of positivity.

The reason younger people have a more negative attitude toward AI? Strong concerns that the technology will eventually take their jobs, reports Bloomberg.

Older people, on the other hand, feel more confident that their experience in the workplace will allow them to hang onto their jobs until they retire.

How is it that the departure of well-known Apple designer Jony Ive undermined departmental resilience so much that former Apple staffers are now accused of stealing trade secrets and covering up the crime?

Forensic evidence

The latest chapter in the ongoing litigation between Apple and OpenAI involves fresh accusations by Apple that former engineer Chang Liu not only attempted to destroy evidence but also created an AI agent to run simulations based on confidential Apple circuit board schematics. 

None of these claims look good, and it’s doubtful Apple would make them if it doesn’t think they can be proved. Apple wants to obtain expedited discovery in the case, which OpenAI claims to be “meritless.” The next big date will be Oct. 1, when Judge Edward J. Davila will need to weigh OpenAI’s denials against Apple’s forensic evidence.

It is interesting that much of OpenAI’s defense seems to revolve around what I call the “cookie jar” argument that it only took information from Apple because Apple’s protections weren’t strong enough in the first place. Perhaps the judge will take the attitude that a loose lid does not justify fingers in the jar, even if it is proved that the lid was actually loose in the first place.

Ternus’ inheritance

That’s the court case, but what Apple’s new CEO, John Ternus, will need to consider as he moves into his new office is what has motivated 400+ former Apple employees — including many from its much-storied design department — to jump ship to its competitor. It’s a challenge he reportedly takes seriously, with recent reports suggesting that restoring the design group back to a cohesive whole is one of his priorities.

He’s going to have to take a very honest look at events; doing so means he will inevitably need to challenge some of Apple’s most sacred cows.

Design a team

Take Ive. There could be a reality in which some of this instability began with his departure. One argument might be that his leadership was what kept the design group together, and once he moved on the cadre collapsed. That doesn’t mean he wasn’t responsible; it was his department and it was his challenge to create a strong succession plan on his departure. Since leaving Apple and forming his own design company, Ive seemed to recruit many of his former staffers, many of whom now work at OpenAI following the sale of his AI company, io, to Apple’s competitor.

It is hard not to see that Apple’s design department post-Ive delivered little success, and while a lazy media class insisted on blaming then-CEO Tim Cook, the truth is that the CDO (Jony Ive was Apple’s Chief Design Officer before he resigned) held his throne long enough to have crafted a more successful approach. Identifying and developing successors is a key task of management anywhere, including in design departments. Problems should be managed down, not managed up.

Safe harbor

His subsequent move at his own company to rehire many of his old team — including Evans Hankey, his anointed successor as design lead at Apple (and co-founder of io), the company sold to OpenAI for over $6 billion — represented the shattering of design team cohesiveness at Apple.  

Another way of seeing it might be that Apple under Cook failed to inspire the design department to such an extent that many of its original members became disaffected and left to work with a competitor. When they did, they found themselves working with their former boss, and his chosen successor as design department lead, Hankey. 

Stepping up

The nature of reality is that one, both, or other arguments may be true. But Apple’s new CEO will need to consider all of them as he plans his next moves. Because while defending Apple’s future in the case against OpenAI is an existential necessity, Ternus must also secure the company’s future by inspiring every department to coalesce around his leadership. That task will include convincing Apple’s remaining designers that the future is better at Apple than elsewhere. That’s going to mean change and it will require a fresh vision for everyone to get behind. 

You can follow me on social media! Join me on BlueSky,  LinkedIn, Mastodon and subscribe to The Core.

Since launching two years ago, Serval has positioned itself as an “AI-native” alternative to incumbent IT service management (ITSM) software vendors, promising to simplify the automation of helpdesk workflows.

Serval’s ITSM platform includes a workflow builder that uses a natural language interface to build code-based workflows. While vendors such as Freshworks and ServiceNow also have workflow automation tools in their products, Serval claims its approach makes automating a recurring task quicker than completing it manually, resulting in more tickets resolved automatically. 

Serval — whose customers now include Spotify, Fox, and Live Nation — has received significant venture funding from Sequoia and others, and was most recently valued at around $1 billion.

With its latest product update, Catalyst — made generally available last week —  Serval promises to take on more of the work needed to build and deploy workflow automations. That includes analyzing ticket histories and other data to identify processes to automate, then configuring settings and connecting integrations to help put those automations into production.

Catalyst also lets admins create long-running background agents that detect problems using historical tickets and other data, proposing fixes before employees even run into an issue, according to Serval. Use case examples include identifying security vulnerabilities, reclaiming unused software licenses, and repairing failed workflows.

Computerworld spoke to Serval CEO and co-founder Jake Stauch about the company’s approach to IT automation and how Serval believes it can differentiate from established ITSM vendors.

The following transcript has been edited and condensed for clarity.

What are the biggest challenges IT teams face when resolving employee requests? “Traditional ITSM tools are built to track tickets; Serval is built to resolve requests. These tools do a very good job of saying that you have work to do, but that doesn’t do anything for the employee. The problem with employee support is really around automation: somebody has to go and build all the automations employees need. 

“Historically, automations have been built in ITSMs as an afterthought, with drag-and-drop workflow builders. You build out these giant flow diagrams; they take months to build and then someone has to maintain them. But what happens when the business process changes, or you’ve got a new approval step for password resets? You’re doing surgery on all these different workflows. 

“Where we differentiate is making automation easier. We think that the best way to build automations is not a drag-and-drop workflow builder, but letting AI do what it does best, which is to write code. Our code generation agent builds workflows for large enterprises to automate not just simple help desk requests, such as password resets, but longer employee journeys, including onboarding and offboarding, job changes, and reporting requirements.”

What are the main components of the Serval platform? And how does Catalyst fit in? “You can think of the system as two agents, though obviously it’s more complicated than this. 

“There’s a help desk agent for employees. An employee can say, ‘I’m locked out of my computer,’ ‘What’s our Wi-Fi password?’ or ‘My laptop stopped working,’ in Slack and Teams, over the phone, or via email — and the agent resolves it right there. 

“We have a separate agent, called Catalyst, that builds those automations. Admins talk to Catalyst, which can build any automation, like password reset workflows, onboarding, offboarding, reporting. Then it helps deploy them either for the help desk agent to resolve employee requests, or for others on the team to use. It even builds background agents that do investigations, pull data from different systems, and come back with reports and propose new workflows.  

“The old Workflow Builder was more limited, in that it translated your prompt into code and created a code-based workflow. You could say, ‘I want a workflow that resets a Microsoft Entra password,’ and it could build that. 

“With Catalyst, you can say, ‘Build a bunch of password-reset workflows for all the different [identity providers] that we have,’ and it’ll build workflows for Okta, Google, and Microsoft, with awareness of what’s already been built. 

“It can also tap into other systems to build dashboards. It can analyze my ServiceNow data and build charts showing all the info, and then propose automations based on its research. It can go a step further and build user interfaces. If you say, ‘I want a form for ordering laptops for users,’ it can build that form, which you can then deploy for equipment purchases across the company. 

What happens if Serval’s AI agents don’t behave as expected? What safeguards are in place to prevent them from taking unintended actions? “Serval’s architecture is built around a separation of powers: one agent builds automations, and a separate agent uses them. Catalyst is gated to admins only and writes deterministic, code-based workflows. The help desk agent that end users interact with can call only the workflows that admins have already published; it cannot create new workflows or reach outside that pre-approved scope. 

“Every workflow carries its own permissions and approval gates that aren’t subject to LLM judgement. This means that while the help desk agent reasons freely over the employee request, any action it takes can be traced to a specific workflow run, and auditors can review both the approvals and underlying code. 

“That combination — of deterministic tools plus an air gap between the agent that builds the tools and the agent that uses tools — is what lets IT teams get the benefits of AI without handing broad system access to a model making judgment calls in real time.”

ServiceNow and other ITSM vendors are also adding AI agents and agent-building tools to their platforms. What is different about Serval’s approach, and what would prevent these firms from replicating it? “The difference isn’t that Serval has AI and legacy vendors don’t. It’s that Serval was built AI-native from day one, rather than layering agents onto two decades of custom tables, business rules, and workflow logic designed for a pre-AI world. That accumulated complexity is exactly what makes even simple automation changes slow and expensive on legacy platforms today; adding an AI layer on top doesn’t make it go away.

“That’s the real barrier to replication. It isn’t that ServiceNow and other incumbents can’t ship agent-building features; several already have. It’s that their existing customers depend on years of customization built on the old architecture, and incumbents can’t tear that out without breaking what those customers already rely on. 

“Serval built the automation engine and the system of record together from the start, with no legacy structure, so building a workflow, enforcing governance, and capturing the data that improves future automations all happen in one place, instead of being stitched together across systems.

“That’s also why many of our customers choose to fully replace their incumbent ITSM tool rather than run Serval as an add-on layer. When Serval is the system of record, not a layer on top of one, every ticket and workflow feeds directly back into better automation suggestions over time.”

Serval can be implemented on top of ServiceNow or another ITSM tool. Can that be a long-term arrangement for customers? “It can be as a starting point – no one wants it that way [longer term]. It may have taken you six years to get ServiceNow implemented, so we don’t expect you to get off it next week. We have to give you a path to get value out of Serval on the day you start.  

“When you begin that way, Serval syncs data back into ServiceNow. You can continue to use ServiceNow — or whatever your legacy ITSM is — with Serval as the automation and orchestration layer on top. 

“While it’s doing that, Serval is also your system of record; it records requests as tickets, and builds out the configuration management database. You end up with two systems doing the same thing, one of which is also automating the work and isn’t 20 years old. 

“Some customers migrate the whole platform immediately, if their ServiceNow renewal is coming up. If their renewal is years into the future, they take a slower approach, using Serval for orchestration automation before the full migration.”

What potential do you see for other enterprise use cases outside of IT? “We don’t have any customers that only use Serval for IT. IT actually accounts for a minority of the teams in Serval today. We talk so much about IT because IT is the buyer of this product of this category; they bring it into the organization and it spreads to other departments, to HR and finance and legal. 

“A typical customer of ours has 13 different departments deployed including IT, security, finance, legal, sales, operations. So, yes, we market towards IT, but IT is really the gateway into the organization.”

What impact does increased automation have on customers’ IT teams? “For a lot of companies, the most technical — and often most expensive –\— resource is the IT team. Yet they spend their day resetting passwords and moving people into groups. It’s a waste of their talent, of their capabilities. 

“Our customers find that when admins no longer have to do that work, they can deploy them across the organization. They can bring AI to different functions, and act as internally deployed engineers that build automations for the rest of the business. You’re enabling them to be builders versus ticket takers, and transforming the IT function from a service function to a building function.”

Do customers reduce IT headcount as they automate more of this work? “What we’re consistently seeing is redeployment, not reduction. Automation frees up admins’ time — in Perplexity’s case [a Serval customer], one to two hours a day per admin — and that time gets reinvested into building automations for the rest of the business: HR, finance, legal, security, and engineering. In a lot of cases, it’s less about IT shrinking and more about IT keeping pace with headcount growth everywhere else in the company without having to add staff of its own.

“Automation doesn’t mean admins step away entirely. They’re still verifying workflows and approving the ones that touch sensitive systems. What changes is that they’re doing that instead of manually resetting a password or provisioning a Slack channel for the hundredth time. 

“What we’re seeing is an unlimited demand for the skills that IT possesses, and so when tickets get automated and IT teams can build amazing tools in Serval, they become even more valuable to their organization, not less.”

Microsoft is working to resolve a widespread Microsoft 365 disruption that has entered its second day, with search and other functions across Exchange Online, SharePoint, OneDrive, Teams and Microsoft 365 Copilot still affected despite the recovery of much of the Exchange Online mail flow.

The incident, according to a report published by the University of Pennsylvania’s IT department based on data from Microsoft’s admin console, began with Microsoft investigating an increase in user reports of Exchange Online problems at 11:55 a.m. UTC on August 31.

By 12:33 p.m. UTC, Microsoft said it had isolated a common failure pattern across affected Exchange Online requests associated with authentication and protocol connectivity, and was working on potential remediation options.

About an hour later, it said that a potential corrective action for the affected infrastructure had been identified and the nature of its deployment was being evaluated.

However, another hour later, it reported that the problem had extended beyond Exchange Online to other Microsoft 365 services, but stopped short of calling it a complete or total outage.

Instead, it listed a range of degraded or failed functions across services, including Exchange Online connectivity and search, SharePoint Online and OneDrive for Business search, file access, synchronization and content loading, Teams search, calendars and presence, and Microsoft 365 Copilot prompts requiring Microsoft 365 data.

A few minutes later, it listed 3:08 p.m. UTC as the official start time of that broader incident and soon attributed the root cause to “an issue within a core authentication configuration used by multiple Microsoft 365 services.”

Microsoft then spent the next few hours testing how to restore the affected authentication components and determine why they were not being deployed as expected. At 4:36 p.m. UTC, the hyperscaler said it was re-examining recent changes and exploring ways to safely restore the components, including potentially reverting an update received by affected infrastructure.

By 5:55 p.m. UTC, Microsoft said mitigation testing had produced positive results and that it was continuing to test and implement strategies to apply the necessary authentication component.

Later at 6:40 p.m. UTC, the hyperscaler began implementing a targeted mitigation to reapply core authentication components across a sample of affected infrastructure while reporting that some users were beginning to see recovery in certain impacted scenarios and that it would incrementally expand the mitigation.

Microsoft subsequently expanded the mitigation and began restarting targeted sections of infrastructure.

Mail flow recovers, but search remains affected

The recovery was not uniform for all affected enterprise customers immediately.

While Microsoft, by late Monday, was reporting widespread recovery of Exchange Online mail connectivity, it warned that some organizations could take longer to drain backlogged mail queues.

The company subsequently said it had validated persistent recovery of mail flow across the environment and shifted its focus to restoring search functionality.

That partial recovery did not, however, bring the broader incident to an end.

Microsoft continued working on search-related issues into Tuesday, restarting affected infrastructure and reapplying the targeted authentication-component fix across affected environments.

In its latest update at 2:39 a.m. UTC, the company said those efforts had “yielded progress,” with incremental improvement in service-health telemetry associated with search functionality across a sample of the affected environment.

It is still continuing the remediation across additional sections of infrastructure but the hyperscaler has not provided an estimated time for full resolution.

It has said that it would provide an estimated recovery timeline when one becomes available, leaving the duration of the disruption uncertain as the incident enters its second day. That for enterprises could delay routine business workflows and leave employees without access to critical information and tools until services are fully restored.