Kategorie

The Matanbuchus malware loader has been seen being distributed through social engineering over Microsoft Teams calls impersonating IT helpdesk. [...]

Google has filed a lawsuit against the anonymous operators of the Android BadBox 2.0 malware botnet, accusing them of running a global ad fraud scheme against the company's advertising platforms. [...]

A novel malware family named LameHug is using a large language model (LLM) to generate commands to be executed on compromised Windows systems. [...]

Cryptocurrency exchange BigONE announced that it suffered a security breach, in which hackers stole various digital assets valued at $27 million. [...]

The Chinese state-sponsored hacking group known as Salt Typhoon breached and remained undetected in a U.S. Army National Guard network for nine months in 2024, stealing network configuration files and administrator credentials that could be used to compromise other government networks. [...]

A critical vulnerability (CVE-2025-20337) in Cisco's Identity Services Engine (ISE) could be exploited to let an unauthenticated attacker store malicious files, execute arbitrary code, or gain root privileges on vulnerable devices. [...]

Over the last decade, Slack has amassed millions of users. The average user has Slack open 10 hours a day and actively uses it a couple of hours daily.

A lot of enterprise knowledge flows through channels, direct messages, and meetings. Extensive new generative AI (genAI) features added in recent years assist in even better communication and collaboration. 

Slack, which is owned by Salesforce, hopes genAI will make its software a “work operating system” where digital labor can also communicate and orchestrate business processes. Some of the newer features include enterprise search, summaries, meeting notes, and translations.

And there’s more cooking. Rob Seaman, chief product officer at Slack, gave Computerworld a glimpse of the company’s thinking and experiments. (This interview was edited for length.)

srcset="https://b2b-contenthub.com/wp-content/uploads/2025/07/Rob-Seaman-CPO-Slack.jpeg?quality=50&strip=all 800w, https://b2b-contenthub.com/wp-content/uploads/2025/07/Rob-Seaman-CPO-Slack.jpeg?resize=150%2C150&quality=50&strip=all 150w, https://b2b-contenthub.com/wp-content/uploads/2025/07/Rob-Seaman-CPO-Slack.jpeg?resize=300%2C300&quality=50&strip=all 300w, https://b2b-contenthub.com/wp-content/uploads/2025/07/Rob-Seaman-CPO-Slack.jpeg?resize=768%2C768&quality=50&strip=all 768w, https://b2b-contenthub.com/wp-content/uploads/2025/07/Rob-Seaman-CPO-Slack.jpeg?resize=697%2C697&quality=50&strip=all 697w, https://b2b-contenthub.com/wp-content/uploads/2025/07/Rob-Seaman-CPO-Slack.jpeg?resize=168%2C168&quality=50&strip=all 168w, https://b2b-contenthub.com/wp-content/uploads/2025/07/Rob-Seaman-CPO-Slack.jpeg?resize=84%2C84&quality=50&strip=all 84w, https://b2b-contenthub.com/wp-content/uploads/2025/07/Rob-Seaman-CPO-Slack.jpeg?resize=480%2C480&quality=50&strip=all 480w, https://b2b-contenthub.com/wp-content/uploads/2025/07/Rob-Seaman-CPO-Slack.jpeg?resize=360%2C360&quality=50&strip=all 360w, https://b2b-contenthub.com/wp-content/uploads/2025/07/Rob-Seaman-CPO-Slack.jpeg?resize=250%2C250&quality=50&strip=all 250w" width="800" height="800" sizes="(max-width: 800px) 100vw, 800px">

Rob Seaman, chief product officer at Slack

Slack

How are you balancing genAI integration with people’s conventional Slack habits?  “We think of AI in two forms — embedded AI and open-ended AI. Embedded AI helps within our existing interface. We automatically summarize PDFs so people don’t have to open files. In our activity feed, we’re scanning activity and extracting action items so you don’t read everything chronologically. 

“We can put assistants on the right-hand side, but I don’t think that’s how it’s going to end up. I don’t think we’ve figured it out yet.

“I don’t think today’s interfaces will hold with AI. You’re seeing search and assistants blend in consumer apps, and that may happen in Slack and enterprise software.

“The question is: how do you seamlessly go from AI search and assistant experiences to colleague communication without jarring context switches? We haven’t figured it out yet, but we’re prototyping internally.”

Can you share other examples of experiments? “I’ll give an example — I don’t know that we’ll ship it, but you could come into Slack and have a prompt bar and not see your sidebar with all channels. That’s a hyperbolic experiment we’ve got. 

“The comments, feedback, and reviews are mixed internally. We’ll do these [experiments] sometimes — and it’s like a pretty big change for some of our staunch users within our employee base.

“You might imagine coming into Slack without your sidebar, just an open-ended prompt that can search, communicate with colleagues, and handle agent tasks. That’s a big change from today.”

Are you thinking about alternatives to text-based chat? “We’re actively working on a feature that turns all your unread messages into a podcast of your chosen length and format. You can listen on your ride back home.

“We’ve done work to reduce meeting overload through asynchronous video and audio clip sharing. Summaries can help, so you don’t have to listen to everything.

“We’re looking at alternative content consumption forms and doing exciting work in text-to-speech.

Are you seeing any hesitation around AI adoption? “A lot depends on the trust the software builds. One is creating value, two is conveying trust. That’s where we spend tons of time. 

“Company adoption follows industry risk curves. Their data isn’t used to train, or it’s not going outside the firewall. Their data is their data.

“We’re working on ‘profile summaries.’ At large companies like Salesforce, I get DMs from people I don’t know. I mouse over — just name and title, which doesn’t help. We added a ‘Tell me more about person’ button. It’ll go off and we’ve prompted it to be positive and look at public channels. It writes a positive CV about their recent work and accomplishments.

“It’s not looking at direct messages and builds trust in the system. That’s what we’re working on — discovery in context of how you already use Slack, but also trust building.

Slack wants to use AI tools to help users learn more about colleagues they don’t know.

Slack

Do you see Slack as a communication layer for agents to talk and collaborate? “Absolutely. I think we have a continued role as a communication layer, but the emerging role for Slack is as the agent command center for users within teams and companies. Tasks are orchestrated and executed by agents on behalf of users.

“I look at AI-native companies for inspiration — Cursor, ElevenLabs, Vercel — they’re putting their agents in Slack because it makes sense. It’s a natural place.

“MCP and agent protocols are very exciting. I think Slack can be an MCP server and an MCP client. We’re working on the server part, but we could be an MCP client for other MCP servers on the desktop or in the browser.”

How do you get users to adopt and understand new AI features in Slack? “I’ll speak openly about what’s been a challenge for us, which is discovery. It’s something we’re spending time on — myself and Ethan [Eismann], our head of design.

“As we add more to Slack, we do not want to mess up the core communication experience. We need users to discover these things in their normal workflow in a non-distractive way that creates value, builds trust, then kicks out to deeper use.

“Gemini does this well — they make recommendations within the product, you try it, then kick out to dedicated Gemini experiences. You discover it, then go into a more dedicated experience.”

Any other cool features coming soon? “There’s one, AI Explain. As a product person, I get pulled into incident channels all the time. When something’s wrong with Slack, there’ll be an incident commander, engineers. I literally don’t know what they’re talking about.

“When I look at messages with acronyms I don’t know, if I mouse over a message, we now have little stars on it. If I click it, it takes that message, all the ones around it, smartly breaks it apart into searches, executes searches, and comes back with an explanation for me.

“That would have been like 10 searches from my past. That’s a great candidate for launching into an agentic experience. I got an explanation, now let’s say I want to take a next logical turn. That’s where we can train users that Slack’s a lot more than just communication.”

With Ai Explain, Slack can help users better understand messages and context.

Slack

How hard can it be to abandon a 10-year old operating system in favor of something more modern, more stable, and more secure? That’s the opportunity IT enjoys with the coming expiration of Windows 10 support in October, giving admins the freedom to make a better choice.

Sure, not everyone in IT is shifting to Apple. Many enterprises remain trapped in the Windows ecosystem through reliance on proprietary and/or third-party applications that do not support the Mac.

There’s also the pernicious nature of the Windows lock-in, which means many in IT have a vested interest in sticking with the platform, even when it doesn’t necessarily provide the best tools for the job.

It’s time to break free

Yet for many, and certainly those in regulated industries in which privacy and security are important, Apple’s platforms do deliver the best toys for the task; you can see this in Apple’s accelerating Mac market share. The fact you can run Windows in VM on a Mac is just a chef’s kiss to support the platform advantage.

For many, the problem is the transition itself. They know that migrating from Windows to Mac will generate some problems — employees will need to re-learn the system, certain software and security routines that are taken for granted will have to be revisited, and there will inevitably be some forgotten apps and services that must be replaced. (To be fair, Apple knows this, too, and continues to improve enterprise deployments across its products.)

That means that any migration to the Mac will take a great deal of planning — or, perhaps, support from one of the many enterprise-focused Apple specialists that exist now, from Addigy to Jamf, Fleet to Hexnode, Kandji to JumpCloud, Mosyle, ScaleFusion, and more.

Each of these firms can help you figure out some of the challenges to Mac deployments at your company, and you’ll find that at least one of these service suppliers can help build up the IT infrastructure you need to manage your new kit.

Be sure before you walk out the door

The big strategic picture is that if you want to make a move Mac-ward, you’re not alone — and if you live near an Apple retail store, you can avail yourself of help and resources provided by the company’s own small army of enterprise tech specialists.

While help is available, the extent of this learning curve might put off some IT decision makers who now need to upgrade their systems from Windows 10. But decision makers should also balance the risks and challenges of upgrading to Windows 11.

What about the impact on operational devices, third-party hardware such as printers or scanners, on-prem apps or even widely used shadow apps that might not have been disclosed? You can rest assured that the larger the scope of the Windows 11 deployment, the higher the associated costs.

Then there are the end users. I’ll be honest, I’ve heard nothing but frustration from users who’ve been forced to move to Windows 11 from 10. Anecdotally, they’ve found that pretty basic tasks (like PDF annotation) are much more challenging on the newer OS.

When you’re managing hundreds of users, small annoyances are consequential time stealers. Each one of those slices of ill-thought-through friction is another support ticket, and while it keeps tech admins employed, these trivial annoyances are budget line items on their own account. We already know that when given the choice, employees will usually opt for a Mac — and forcing them to use Windows software that doesn’t work like it should is not how to make your teams more productive. 

Life goes on

Accepting then that moving to Apple poses problems, and that migrating to Windows 11 also poses problems, IT should take a look at the Total Cost of Ownership between Mac and Windows. If they do, they’ll find that while Mac might seem slightly more expensive to purchase (but even then, not for long), the products last longer, require less tech intervention, are easier to manage, use, and run, hold value far longer, and have longer usable lives.

The modern Mac is also hugely performant, offering amazing computational power in every available machine. The current MacBook Air, for example, delivers far more computing power than the first M-series MacBook Pros. In other words, when you add it up, the Mac is likely to be more affordable than Windows. 

With so much to drive the argument and so little against it, it is of no surprise at all that the Windows 11 upgrade mania is translating into major Mac market share increases, according to IDC, Gartner, and Canalys.

Quite simply, Apple now seems to offer the best choice.

You can follow me on social media! Join me on BlueSky,  LinkedIn, and Mastodon.

If you've worked with Clonezilla''and let's be honest, most sysadmins have''then you know its strengths. Reliable, fast, and no-nonsense disk imaging.

Let's talk about CVE-2025-6558 ''the latest zero-day vulnerability in Google Chrome. If you're managing Linux systems or handling infosec at any level, you should care about this one. It's actively being exploited, which means attackers aren't waiting for you to patch''they're already leveraging it to break into systems.

Large language models (LLMs) such as GPT-4o and Google’s Gemma may appear confident, but new research suggests their reasoning can break down under pressure, raising concerns for enterprise applications that rely on multi-turn AI interactions.

A study by researchers at Google DeepMind and University College London has revealed that LLMs display a human-like tendency to stubbornly stick to their initial answers when reminded of them, but become dramatically underconfident and prone to changing their minds when presented with opposing advice, even when that advice is incorrect.

“We show that LLMs – Gemma 3, GPT4o and o1-preview – exhibit a pronounced choice-supportive bias that reinforces and boosts their estimate of confidence in their answer, resulting in a marked resistance to change their mind,” the researchers said in the paper. “We further demonstrate that LLMs markedly overweight inconsistent compared to consistent advice, in a fashion that deviates qualitatively from normative Bayesian updating.”

“Finally, we demonstrate that these two mechanisms – a drive to maintain consistency with prior commitments and hypersensitivity to contradictory feedback – parsimoniously capture LLM behavior in a different domain,” the researchers added.

The findings challenge conventional assumptions about LLM reliability in multi-turn enterprise applications, where decision support and task automation increasingly rely on AI-generated confidence.

This hypersensitivity to criticism and poor calibration under pressure could introduce hidden risks for enterprises deploying conversational AI in regulated, high-stakes, or customer-facing workflows.

Threats to enterprise AI

Analysts say the tendency of LLMs to reverse their answers under pressure is not a one-off glitch but a structural weakness in how these systems handle multi-turn reasoning.

The DeepMind study reinforces what has been observed in real-world use: models that initially give correct answers often discard them when faced with confident user input, even if that input is incorrect.

“This trait, termed ‘sycophancy’ by Stanford researchers, arises from an overemphasis on user alignment over truthfulness during model fine-tuning,” said Sanchit Vir Gogia, chief analyst and CEO at Greyhound Research. “In enterprise applications like customer service bots, HR assistants, or decision-support tools, this deference creates a paradox: the AI appears helpful while making the system less trustworthy over time.”

As AI becomes embedded in core workflows, organizations are being urged to shift from single-turn validations to treating dialogue integrity as a critical and testable aspect of system performance.

Dealing with nuanced sycophancy

Researchers noted that the sycophantic behavior of LLMs may be partly explained by how they are trained using reinforcement learning from human feedback (RLHF), a technique intended to align responses with user preferences.

However, the study found a more nuanced pattern than simple sycophancy.

While sycophancy typically involves a symmetrical bias toward both agreeing and disagreeing input, the models showed a stronger sensitivity to opposing advice than to supportive input, with the likelihood of changing answers influenced by the model’s initial confidence.

“While this behavior may enhance perceived helpfulness in consumer settings, it creates systemic risk in enterprise environments that rely on AI to enforce boundaries,” Gogia said. “Whether in banking KYC, healthcare triage, or grievance resolution, enterprises need AI systems that assert truth, even when the user insists otherwise. Sycophancy undermines not only accuracy but institutional authority.”

Enterprises will need to adopt alignment strategies that prioritize factual accuracy over user satisfaction, especially when the two are at odds.

The next browser revolution — this time with an AI interface — is just around the corner. At least, that’s how OpenAI, The Browser Company, and Perplexity see it. 

These companies are all developing AI browsers designed to do an even better job of presenting information available on the web.

Perplexity’s Comet is among the earliest such browsers. And once users latch onto it, using Google’s Chrome will feel like using a Walkman. (That’s how Perplexity sees it, anyway.)

To cut through the hype, Computerworld tested an early version of Comet to understand how AI could change browsing in much the same way Netscape shift the online landscape three decades ago.

Comet, which is only available for now by invitation only, uses AI’s best attributes to cut down the manual slog of clicking through links, creating tabs, finding files, and completing web-based work. The centerpiece is a built-in assistant, where users can type in search queries or ask the browser to complete tasks. A chatbot pops up in a side panel of the browser, while the main window shows web pages.

The goal was to compare how an AI-powered web browser and a generic browser would complete the same task. I tested an early version of Comet, with Perplexity promising to  add more functionality over time.

Writing emails

First, I connected Comet to my Gmail account; that allowed the browser chatbot to list and summarize my latest emails. The chatbot was also able to auto-compose and send an email from the Gmail account and it could search through emails by broadening the context beyond just keywords. (Typical Gmail searches look for specific words.)

The AI browser saved time and effort from the conventional click-through process of composing and sending email.

Automatically create tabs

Even better was Comet’s ability to automatically create multiple tabs based on queries to the AI assistant.

Using this query, “Keith Shaw’s Today in Tech — create tabs in Comet browser for last 5 episodes” delivered as promised. It automatically created five tabs linking to Apple Podcasts of the most recent Today in Tech episodes, which is hosted by my Foundry colleague Keith Shaw.

Comet browser used AI to create five browser tabs linking to five recent episodes of Today in Tech.

Agam Shah/Foundry

The AI assistant used reasoning to determine the most recent episodes and browsed the web to find them. The automation saved time compared to the conventional browsing experience of manually determining the last five episodes and finding links via Google search.

Create a shopping cart with one click

Similarly, the query “Create a shopping cart for enchiladas for 1 person on Instacart” automated the creation of a shopping cart on Instacart. It determined that the enchilada was for one person, identified six ingredients, browsed nearby stores, and automatically created a shopping cart. (The checkout and use of coupons required a manual process.)

The AI assistant also followed up by providing directions on how to make enchiladas with the ingredients purchased on Instacart. A typical browsing experience would require knowing the ingredients and browsing through multiple stores to see whether the items were in stock. The automated experience using Comet was a real time saver.

The Comet browser AI assistant filled up an Instacart shopping cart with ingredients for enchiladas.

Agam Shah/Foundry

Automating productivity

Comet’s AI features can also summarize, edit, and suggest improvements to documents. A few weeks ago, my martial arts instructor asked me to document the belt syllabus of Nintai, an eclectic martial art with cross-training in Korean and Japanese styles. 

Creating the syllabus is proving to be a highly involved manual task and includes putting together information scattered through different files. I wanted to see whether Comet would help me automate the process.

I loaded the brown-belt syllabus from a document in Google Drive in the browser. The AI was able to summarize the text and provide suggestions to modify the document. 

Furthermore, the suggestions drew context from references to Nintai and my instructor in my browsing and email history. The AI also summarized videos about the style, which improved the document even further. 

I also asked the Comet chatbot to find documents related to Nintai in my Google Drive. Comet made an attempt, and it was fascinating to see how AI automated the entire browsing process.

The browser window showed in real-time Comet trying multiple ways to access and open files to determine the content, which would help locate files related to Nintai’s syllabus. It couldn’t open the files, but it provided a hint at how AI browsers in the future could understand documents and boost productivity.

Final thoughts

Comet isn’t perfect and retains the traditional manual browsing experience (because it’s built on Chromium). It did take me a while to get used to a AI-assisted browsing and filling up shopping carts, much like the first time I sat in a Waymo. But Comet is still in beta and will likely mature.

OpenAI has teased a possible web browser of its own (the latest speculation is it’ll be called “Aura” and arrive this summer), and is already planning an AI-powered productivity suite.

Meanwhile, The Browser Company’s Dia is available for download for Mac users on a waitlist. Dia was introduced after the company ditched its conventional Arc browser, which was known for its slick design but deemed irrelevant in the AI era.

Who knows what will happen with Apple’s Safari, given the company’s slow start on AI? A Gemini add-on isn’t available for Chrome, and Firefox, Chrome and Opera will also need AI extensions. But it seems likely that conventional browsers without AI features won’t survive what’s coming.

** Nezabezpečená rádiová komunikace umožňuje na dálku ovládat brzdy vlaků ** Bezpečnostní experti na tuto vážnou chybu upozorňovali úřady třináct let ** Modernizace a oprava celé sítě potrvá ještě několik dalších dlouhých let

UK retailer Co-op has confirmed that personal data of 6.5 million members was stolen in the massive cyberattack in April that shut down systems and caused food shortages in its grocery stores. [...]

A 21-year old former U.S. Army soldier pleaded guilty to charges of hacking and extorting at least ten telecommunications and technology companies in the country. [...]

Luxury fashion giant Louis Vuitton confirmed that breaches impacting customers in the UK, South Korea, and Turkey stem from the same security incident, which is believed to be linked to the ShinyHunters extortion group. [...]

To quash speculation of a cyberattack or BGP hijack incident causing the recent 1.1.1.1 Resolver service outage, Cloudflare explains in a post mortem that the incident was caused by an internal misconfiguration. [...]

For a while now, OpenAI has been rumored to be working to develop its own web browser, with the aim of challenging Google’s dominance in the segment. And that browser will apparently be named “Aura”, according to Bleeping Computer. The browser name was referenced in the code used by ChatGPT.

The interface in Aura is said to be reminiscent of ChatGPT — in other words, the focus will be on conversations rather than the traditional clicking of links. Alongside ChatGPT, Aura is expected to offer built-in support for the SearchGPT search engine and the AI agent Operator.

The launch of the new browser is expected to take place later this summer, possibly by the end of July.

A threat actor has been deploying a previously unseen malware called OVERSTEP that modifies the boot process of fully-patched but no longer supported SonicWall Secure Mobile Access appliances. [...]

Multiple Fortinet FortiWeb instances recently infected with web shells are believed to have been compromised using public exploits for a recently patched remote code execution (RCE) flaw tracked as CVE-2025-25257. [...]