Computerworld.com [Hacking News]

It’s cheap to set yourself up as a hacker these days using AI. Someone attacked 105 online retailers over a period of five days, compromising 27 of them — all for an average of $25 per attack, according to research by Israeli security company Gambit.

But the attacks have been going on for much longer. Whoever is responsible used open-source AI harnesses to mount the attack. Gambit has identified three of these: Strix for vulnerability search, Cairn for autonomous end-to-end exploitation, and Hermes to orchestrate the campaign. The attacks have proved to be highly lucrative, with 600,000 active credit card details taken from just two businesses, the installation of card skimmer scripts at five more, and some level of access to an unspecified number of major companies.

It has been a highly efficient attack, most access taking just a few hours and at a minimal cost. OpenRouter was used for AI model access. The capture of the account balance on August 25 showed the attacker spent just $7,005 over a four-week period. This worked out at roughly $25 per attack, with costs ranging from $3.13 for the cheapest target to $79.31 for the most expensive.

Gambit has contacted all the companies concerned, but warned that the intensity of the attacks shows how AI is transforming cyber criminals’ activities by providing a level of sophistication that humans would find more challenging to muster. We are already seeing a new level of incursions to businesses and can expect to see more in the future.

This article first appeared on CSO.

Here’s an interesting concept: an AI company that fires people for using AI. It sounds like a strange way to run a company but there’s a real reason behind it.

OpenAI has been hiring contractors who have been tasked with reading ChatGPT users’ prompts to help improve responses by providing some real human input.

Unfortunately for OpenAI, it found many of them were using AI to train the AI, so not providing a human touch at all, according to a report in 404 Media. The company has subsequently fired many of the contractors, although 404 Media didn’t reveal the number.

None of them can say they weren’t warned, however. The terms for the contractors are set out in their working conditions. “Do not use AI detection tools, or AI yourself. Do not use GPTZero or any other AI detection tool. They are not reliable. Reviewers may not use AI either, including Grammarly and AI translation, to review, write feedback, or write comments.”

Despite this stark warning, many of the contractors turned to AI to assist in the work. OpenAI is very keen to avoid “model collapse,” a phenomenon in which AI models trained on text written by previous generations of AI models perform worse than their predecessors. It’s a form of digital inbreeding that industry observers have previously warned about, and which could have a deleterious impact on business.

One contractor told 404 Media that it was a common practise. “It’s pretty much the one thing that will get you kicked off ASAP. In a group of thousands, there are tons that have been caught.”

It will be hard to ascertain just how widespread the practise is as OpenAI declined to comment on the situation.

Microsoft’s Copilot “super app” has arrived, two months after CEO Satya Nadella confirmed it was in development. It combines conversational Chat, the autonomous Cowork agent, a redesigned coding environment named Code, and Autopilot, the persistent agent formerly known as Scout, into a single Copilot experience.

The idea is to make those capabilities look less like separate tools and more like different ways of getting work done through a unified experience. The end goal is to enable users to describe what they want to accomplish without choosing a mode themselves, and with Copilot routing the task to Chat, Cowork, Code or another capability, Microsoft said.

The individual pieces of this experience are not entirely new, and build on launches and consolidation moves that Microsoft has made over the past several months.

Nadella first described the unified Copilot architecture during an earnings call in July. He said it would bring chat, agents, and business workflows into a single AI workspace while separating the orchestration layer, context, and memory from any one AI model. Microsoft subsequently began unifying its consumer and Microsoft 365 Copilot apps ahead of the broader overhaul.

But Microsoft laid much of the groundwork in the period after its November 2025 Ignite conference, with the launch of Cowork for delegated multistep work, Scout for persistent agentic tasks, and natural-language app-building capabilities that have now evolved into Code.

The broader change goes beyond bringing those Copilot experiences together. With Friday’s updates, Microsoft is also adding components and infrastructure aimed at turning Copilot from an AI assistant embedded across Microsoft 365 into a broader orchestration layer for enterprise work.

These include Fabric IQ and Work IQ for enterprise context, Copilot Managed Runtime for running AI-built applications, a unified plugin registry for connecting capabilities, and usage-based billing and FinOps controls for managing agentic workloads and AI spending.

The inclusion of Fabric IQ in Chat and Cowork, along with Work IQ’s access to data and workflows in Dynamics 365 and Power Platform, will make Copilot more effective for enterprise tasks by giving it a broader understanding of an organization’s data and workflows, potentially reducing hallucinations and improving productivity, said Abhishek Satapathy, principal analyst at Avasant.

“The practical shift is from AI that knows information to AI that can connect what happened, why it happened and what needs to happen next. For example, a business insight can increasingly become the starting point for analysis and follow-up action rather than requiring the user to manually move that context between reports, documents, conversations and business systems,” said Satapathy. “This is particularly relevant for analysts, managers and operations teams working across functions.”

Fabric IQ’s integration with Chat and Cowork is now generally available. Microsoft plans to integrate it with Code soon through its Frontier program. Work IQ will enter public preview over the next month, it said.

Productivity is also the focus of Microsoft’s redesigned Code environment.

It “broadens the audience for software creation” by addressing the middle ground between traditional development and simple automation, bringing more use cases into the scope of what business users can build, including those that may be too narrow for a formal development project but too complex for basic automation, Satapathy said.

That means more internal applications, dashboards and workflow tools can potentially be built without going through a traditional development project, while developers can use the same environment to accelerate application creation and focus on more complex engineering and production requirements, he added.

Alongside Code, Microsoft is introducing Copilot Managed Runtime, a governed hosting environment for applications built with Code, Cowork, and Copilot Studio. The runtime allows those applications to run within an organization’s Microsoft 365 environment while giving IT control over security, governance and access to live data, Microsoft said.

For enterprise developers and IT teams, the runtime could provide a common production layer for AI-created applications, helping enterprises move applications from experimentation and prototyping into production without having to create a separate operating model for each one, Satapathy said.

It could also help IT teams and CIOs curb shadow IT and application sprawl, said Manoj Chandra Jha, principal analyst at Nord-IQ Research. In its governed environment, IT departments can manage applications and agents created by employees consistently for security, identity, policies, lifecycle management, observability, and deployment.

The new Copilot Managed Runtime is currently in public preview and will also be accessible inside the redesigned Code experience, Microsoft said. That experience will roll out to Microsoft’s Frontier program at the end of the month, with broad availability in the coming weeks, it added.

Beyond productivity, faster scaling of AI applications and curbing application sprawl, Microsoft is also giving enterprises more tools to manage the cost of agentic AI by expanding its cost-management capabilities in Agent 365 to include Code and Copilot Managed Runtime.

These capabilities will allow administrators to manage AI spending policies at scale, control access to model families and route credit requests through existing approval workflows, while giving visibility into the outcomes generated by agentic workloads to decide where to sustain, expand or optimize usage, the company said.

End users will also be able to view their credit usage, remaining balances and usage history directly in Copilot, helping them manage their AI usage, it added.

For Bhupendra Chopra, chief revenue officer at IT consulting firm Kanerika, the expanded FinOps capabilities will be critical for CIOs, particularly for understanding if agentic workloads and automation are delivering sufficient business value and if they need to be reworked.

“These capabilities can reveal correction rates for agents. If employees regularly have to undo an agent’s work, that can indicate stale context, poor business rules or an unreliable data foundation before it shows up as a larger operational problem,” Chopra said.

“These capabilities can reveal another insight: whether the time supposedly saved by automation actually translates into a measurable business outcome. Enterprises can automate 20 minutes of work but simply spend that time elsewhere, meaning the productivity gain only matters when it produces a measurable result,” he added.

Alongside these FinOps capabilities, Microsoft is also extending usage-based billing to more of the capabilities being brought together under Copilot. The model is not new to the platform, having been introduced for Cowork when the agent became generally available in June.

While Cowork, Code and Autopilot, along with new long-running agentic capabilities and frontier models, will use usage-based billing, Chat and Copilot across Microsoft 365 will remain covered by the user subscription license, the company said.

That distinction could require CIOs to think about AI spending differently from traditional productivity software, Satapathy said.

Instead of budgeting primarily around the number of users with Copilot licenses, CIOs may need to consider what work is being delegated to agents, how much that work consumes and what business value it produces, he added.

“Two employees with the same Copilot license, for instance, could generate very different AI costs depending on whether one uses Copilot for assistance while the other delegates longer-running work to agents,” he further said.

However, that might still be some way off, with several of the key capabilities still in preview or yet to roll out.

Home, the new starting point in Copilot where Chat and Cowork come together, will begin rolling out through Microsoft’s Frontier program in the coming weeks, while Autopilot is expanding to private preview at the end of the month, Microsoft said.

Adobe is preparing for a future where users increasingly interact with its software through AI assistants, rather than always going directly to its applications.

That’s the idea behind its expanded integration with Anthropic’s Claude, as well as the first connector for Google’s Gemini, announced Thursday.

The updated Claude integration adds new ways for users to interact with Creative Cloud applications via text prompts, with 80 “tools” available across applications such as Acrobat Photoshop, Premiere, Express, and Lightroom. That’s up from the initial 50 at launch in April.

The Adobe plugin lets users take actions across various Creative Cloud apps, such as retouching images or resizing videos for social media posts. The latest additions include the ability to convert PDFs to different file formats.

In addition, Adobe has introduced interactive editor tools available in Claude. These offer greater control over Express and Acrobat files, Adobe said.

In Acrobat, it’s possible to edit text, for instance, reorder pages, or add comments and highlights. For Express, this means the ability to select and customize fonts, colors, and more.

The integrations could broaden access to Adobe’s software among workers “who never fully learned how to use Adobe’s tools,” said Keith Kirkpatrick, research director at The Futurum Group.

The ability to control elements of Creative Cloud apps via prompts “removes the learning curve, and can potentially help speed up the time it takes to go from an idea to a mostly finished output,” said Kirkpatrick.

Professional designers, meanwhile, will be “freed up to work on finishing ideas, rather than the spending valuable time doing the initial ideation and refinement,” he said. “It should also help them focus on only the highest-value work, instead of the routine or mundane projects.”

The Claude connector is part of a wider Adobe strategy to make Creative Cloud applications available within leading AI assistant tools. This includes similar integrations with OpenAI’s ChatGPT, Salesforce’s Slackbot, and, as of this week, Google’s Gemini.

AI usage by a range of workers is one the rise, even if regular adoption is still limited. Around 15% of US employees use AI daily in their role, according to a Gallup poll from May this year, while 30% use it a few times a week or more.

Gartner predicts that spending on enterprise AI assistants will rise from $17 billion in 2025 to $71bn in 2030, while knowledge workers are forecast to spend more time interacting with these tools in their daily work than with any other application over the next three years.

For Deepti Pradeep, Adobe’s senior director of agentic AI, the emergence of AI assistants creates another way for customers to access Adobe’s tools, rather than a replacement for its own applications.

“I think you will find a mix of all types of users; people who are in LLMs doing their tasks, and people who want to have more control and creativity as they’re going deep in our flagship apps,” she said.  

“Our belief is that we want to be wherever the user is, whether they’re in the flagship apps ‘in the zone’ doing their thing, or whether they’re outside and they want to accomplish the task. So I don’t think it’s ‘this’ or ‘that’ — I think it’s everything. And we just want to be valuable in every moment that the user is in.”

Going forward, Adobe also plans to extend the interactive editor controls available now available for Acrobat and Express to a wider range of apps such as Photoshop. However, the emphasis will be on more streamlined interfaces, with more complex functionality remaining in the core application.

“With image editing there’s so much one can do, and we have very comprehensive products where we possibly can’t bring all that [to AI assistants such as Claude],” said Pradeep.

“So we’re trying to figure out what’s the right level that makes sense for the user in that given context. We want to ensure that we’re bringing relevant, value-added things. What we don’t want to do is over-engineer it.”

Adobe for Claude includes new interactive editing tools for Acrobat and Express.

Adobe

Liz Miller, VP and Principal Analyst at Constellation Research, said there’s a lot of “buzz” from software vendors about the “start of an age where you never need to log into an application’s interface,” she said. This is exemplified by Salesforce CTO Parker Harris’s comments that customers may not need to log in to the CRM app at all in the future to get work done.

For some workers that may be accurate, said Miller, and the ability to access Adobe apps via third-party AI assistants “will 100% change the space and place they choose to work.”

But, just as access via mobile didn’t kill off Photoshop, “working in Claude will be a new way to work in, with, and through, Creative Cloud,” said Miller. “This is more about the future styles of work than it is about a SaaSpocalypse. Adobe is rightly identifying a new work surface and empowering creators to work where they choose.”

Miller said she already sees demand for Adobe’s integrations in third-party apps, “especially from non-creative, more technical users where updating a technical document that exists as a PDF was once a time-consuming task.

“A prompt-based update leveraging Acrobat and Express integrations with Claude fits within that individual’s workflow,” Miller said.

Adobe doesn’t provide daily or weekly usage statistics around the use of the existing Claude integration. It’s still early days in terms of uptake, the company said, though there is strong repeat use among early adopters and positive qualitative user feedback.

Some have questioned whether wider use of enterprise AI assistants could negatively impact software vendors such as Adobe, which announced a $25 billion share buyback program in April amid investor concerns about the threat AI poses to its business.  

Miller said that making Creative Cloud apps available across tools such as Claude could be beneficial for Adobe, widening the reach of its own products.  

“This could be an interesting opportunity since Claude is not replacing Adobe’s capabilities or applications in this scenario,” she said. “In fact, it more closely connects Adobe applications to work being done using Claude.”

Kirkpatrick sees the integrations as a way to make Adobe’s creative tools available to a wider audience of office workers.  

“I view Claude and Gemini as an easy-to-use front door, which expands the reach of the platform to users that might ordinarily not ever really use Adobe products,” he said.

There are three levels of access to the Adobe plugin for Claude.

Guest users require no authentication but can use only a smaller subset of Adobe tools, while those that log in to an Adobe account can access all functionality but with certain restrictions on usage. Paid Creative Cloud users effectively have unlimited access to available Adobe tools in Claude.   

“The actual functionality for guest users, where Adobe doesn’t have basic user identification data, is still limited, which ensures that there’s a reason to further engage with Adobe and its products,” said Kirkpatrick.

Accessing Creative Cloud apps doesn’t require any additional fees or draw on any Adobe AI credits. Adobe declined to comment on future plans to monetize the integration.

For Gemini users, free and paid users can access Creative Cloud tools, but there’s no guest access.

Kirkpatrick describes Adobe’s approaches as a form of freemium model, with more functionality unlocked as users take steps such as signing up for a free Adobe ID or purchasing Creative Cloud or Firefly subscriptions.

This chimes with Adobe’s wider strategy for driving adoption of its own AI tools, with CEO Shantanu Narayen stating during a recent earnings call that attracting new users matters more than “short-term relief” around pricing.

“In a world where new tools and technology come on the market frequently, driving adoption through ease-of-use and then backing that with enterprise-grade security, IP-rights management, and scalability, Adobe’s freemium model appears to be the right play,” said Kirkpatrick.

On October 1, Google’s first test satellite for Project Suncatcher is set to be launched. This is the next step in the company’s plan to build AI data centers in Earth’s orbit, according to The New York Times. The idea is, among other things, to harness solar energy in space instead of building ever-larger data centers on Earth.

The first satellite, MVP, is about the size of a refrigerator and was built by the satellite company Planet Labs. On board are four of Google’s TPU accelerators, which will be used to run Gemini in space.

MVP is equipped with solar panels that generate about one kilowatt. This, combined with cooling limitations, means the TPU chips can only run for 15 minutes at a time during testing.

Google plans additional launches in 2027. The goal is a network of AI satellites that communicate with each other via high-speed laser links. However, Google expects it will take several years before Project Suncatcher can be developed into an actual product.

Data centers on Earth:

• California joins US states clamping down on data center gold rush
• Data center backlash could slow CIOs’ AI plans
• Can Microsoft build massive AI data centers and meet climate goals without ‘greenwashing’?
• Is the rise of genAI about to create an energy crisis?

I once had a friend who gave $1,000 to an online fraudster who claimed to be a tax assessor. My own father (bless his soul) once managed to hand over several hundred pounds to a lawyer from Africa promising to send him funds. 

We know it happens. 

We know people impersonate trusted entities to trick us into giving them our money. It’s a scam, one that’s taking place at every level of digital existence. 

Apple is here to help.

Yet another great feature new to iOS 27, Apple’s new scam-prevention tool is a welcome intervention, giving individuals and businesses another line of defense against social engineering scams. While it’s not perfect, it is here, and every Apple user — particularly those working in or managing regulated industries, or any business at all that feels it may be a target for scams, fraud, or socially-engineered crimes — should take a look. According to most security reports, that effectively includes all of us. 

On its support site, Apple explains several hypothetical attack scenarios the feature is designed to protect against, such as when an attacker convincingly poses as a government agency or some other trusted entity to trick you into making a payment or changing account or login details. 

These things are very hard for traditional security protection to spot, in part because you are choosing to take the action — security can’t tell you’ve been tricked. Apple’s new system works to bridge the gap. It looks at information it knows about your device and your Apple Account and tries to spot when you’re being scammed, delivering its own risk assessment.

While Impersonation Risk Detection is a new system-level tool, it does require user consent to permit supported apps to ask the operating system for a risk assessment if the user does something that could be in response to a scam. 

That assessment takes the form of a risk level that helps the app decide what to do next. This will usually trigger additional security steps on the part of that third-party app, including a request for identification, a delay in a transaction, or a warning to let you know you might be being scammed.

Supported apps may request one of these risk assessments when you do things like make a payment or change security settings for your account. Apple tells us these assessments come in the form of one of three levels: Unknown, Medium, or High, with High meaning significant signs of suspicious activity have been detected. It is important to note that while Apple raises these flags, it’s the app that makes the decision on what to do next. Apple’s part ends once it raises the warning.

The settings for the feature also let you take a look at which apps have requested a risk assessment, and why. At least, it will once the feature is activated and has been running for a while. You’ll find a list of requesters in Recent Activity, while the Reasons for Access will show you what actions prompted the request. 

If you do choose to share information with developers, they will only get the scam warning signals generated using information about your device and Apple Account. Apple, however, will learn the type of action you attempted in the app, and may also learn relevant details such as the number of calls or emails you’ve made. 

This is what Apple says it checks:

• On the device, Apple analyzes interaction patterns, timing, context, and basic sensor data to generate the risk level. 
• Apple never analyzes the content of your Photos, Messages, or Mail.
• Apple learns the type of action you attempted in the app only when the app requests the risk assessment.
• Apps receive only the risk level, not the data.

Because the tool requires that some of your information is shared, Apple has left Impersonation Risk Detection off by default. Here’s how to enable it or just take a look:

• Open Settings > Privacy & Security.
• Scroll down the pane to find the Impersonation Risk Detection section, which sits beneath the App Advertising item. 
• Tap this, and on the next page you’ll find a toggle to share information with developers of apps that support the feature. 
• Toggle this to on for the protection to kick in. Apple warns it may take four hours to come into effect. 

Once approved, compatible apps can request Apple’s Impersonation Risk Detection signals to see if your device or account shows signs that you’re being scammed. And hopefully you — or my dad — won’t get fooled again.

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

The 7,000 km (4,350 mile) cable, called Petal, will connect the US and France. Meta is partnering with NEC and Sumitomo Electric Industries to build it, and working with French telecommunications company Orange to coordinate the landing of the cable on the French coast.

Petal will have more than double the capacity of today’s fastest transoceanic cables. Last November, AWS laid plans for a 320 Tbit/s cable between the US and Ireland, while a cable connecting Australia and the US, announced in January this year, will deliver 400 terabits per second.

The need for greater data connectivity has prompted a boom in subsea cable deployment, but their existence is making some US lawmakers rather twitchy, as they are fearful of the security implications. There have been calls for the big service providers to provide details of all the cables they have in place. The concerns are warranted given the number of incidents involving subsea cables.

This article first appeared on Network World.

Google’s Gemini 4 AI model is in the early days of post-training, the phase in which a base AI model is refined to behave reliably, and should be released “much earlier” than the end of this year, Google DeepMind head Koray Kavukcuoglu told The Information at its AI Agenda Live Summit. Some Google observers have speculated that this could be as early as October.

Kavukcuoglu recently replaced DeepMind founder Demis Hassabis as head of the Google business unit.

The launch of Gemini 4 may lay to rest concerns about the delayed release of Gemini 3.5 Pro, which Google was originally expected to announce at its May 2026 developer conference.

While less capable models in the Gemini 3 family have been frequently updated, Gemini 3 Pro has only been updated once since its November 2025 released. In contrast, OpenAI — spurred on by Sam Altman’s “Code Red” memo — has released three updates to its frontier AI model since then: GPT 5.5 Pro, GPT 5.6 Astro, and GPT 6 Astro. Anthropic, too, has updated its most powerful Claude model several times.

Google has not been entirely idle in the AI arena, concentrating its efforts on updating less powerful, more affordable Gemini versions. In July, it announced three Gemini Flash models, aimed at more routine AI tasks, but the company remained silent about its high-end alternative.

AI PCs that cut token costs? That may appeal to enterprises.

A new breed of agentic AI PCs promises to do exactly that. The powerful laptops can cut token costs by completing AI work locally instead of sending it to expensive LLMs in the cloud.

HP’s newly announced ZBook Ultra G3a mobile workstation and upcoming laptops from other PC makers have powerful GPUs and large memory pools designed for AI and graphics. The ZBook Ultra G3a uses an AMD Ryzen AI Max Pro processor, while laptops from other PC makers will feature Nvidia’s RTX Spark superchip.

Nvidia’s Blackwell GPUs have dominated data centers, but the company is now targeting a larger chunk of the PC market with the RTX Spark, which includes a Blackwell RTX GPU. The first wave of RTX Spark laptops coming from Asus, Dell, Lenovo, and other PC makers is targeted at consumers, with enterprise models planned for later. HP’s ZBook Ultra PC, on the other hand, is squarely aimed at enterprises.

Without connecting to the internet, the PCs can run agents and chatbots, generate video, and write code on AI models with billions of parameters.

“It’s the only PC where you have enough memory to run the large models, enough compute to run them fast, and at the same time, all of your tools that we’ve accelerated for years… are all working,” said Gerardo Delgado, senior director of product management at Nvidia.

To be sure, cloud-based AI models are more powerful and capable. But AI in the cloud is expensive, and these PCs can take some of the heat off those bills.

“People are starting to get either burned or really concerned when they’re starting to get their AI token bills,” said Brian Allen, manager for Global Z workstation products at HP.

At least 20%-25% of high-end AI workloads in the next two to three years will be running on AI PCs versus totally in the cloud, said Jack Gold, principal analyst at J. Gold Research.

The agentic AI machines won’t be inexpensive, but they will cut recurring cloud costs, which is especially important in the ongoing research and engineering work that produces multiple versions of a project, Gold said.

“We will continue to see more offerings in this space in the next six months, particularly as more vendors release chips specifically targeting this space,” Gold said.

Early AI PCs, notably Windows machines with Microsoft’s Copilot+ branding, offloaded some basic assistant tasks, but the ZBook Ultra G3a is an AI mobile workstation aimed at high-end enterprise workloads.

“I’m actually building large language AI models. I’m doing fine tuning. I’m doing inferencing,” Allen said.

For example, designers can use AI in Autodesk Revit 3D modeling software at an airport without an internet connection. That is made possible by the inclusion of Perplexity’s Portable Computer offering.

Designers can then connect to frontier models in the cloud to advance the project. Perplexity taps across roughly 19 frontier models to find the right fit.

Model Context Protocol (MCP) servers bridge the AI models with applications on the PC with a user’s permission. AI on PCs can keep sensitive data within the company’s boundaries.

HP has partnered with Perplexity on Revit, but “you’re going to see those MCP connectors really start growing and tying into more software applications,” Allen said.

HP’s laptop, expected to be available in October, has packed up the CPU, memory, and AMD’s GPU tightly on one chip, so data moves quickly between the components. That’s important for AI and different from regular PC designs that offer upgradable memory.

HP hasn’t shared pricing for the mobile workstation, but Gold said it will be far costlier than regular AI PCs such as Microsoft’s recently upgraded Surface Laptop 13-inch and Surface Pro 12-inch models with Qualcomm’s Snapdragon X2 Plus chips, which start at $1,199 and $1,149, respectively.

HP provides an “ROI calculator” that lets curious shoppers play around with the cost of running AI locally versus the cloud-based cost. “You can sort of say, ‘hey, this PC pays for itself in nine months based upon this type of usage,’” Allen said.

ROI calculators like HP’s “have a number of assumptions built in that may not be applicable to every workplace,” said analyst Gold. “But calculators like these are still valuable for companies to do some higher-level evaluation of options.” In some cases, enduring high cloud-based token costs could be worth it for complex workloads and bringing products to market quicker, he said.

Nvidia received an intense wave of customer requests for mobile workstations after models surpassing Opus 4.6 excelled at coding and rocked the AI world. “That is the poster child right now, in development teams, coding teams, moving really quickly into local hardware,” Nvidia’s Delgado said.

Delgado said agents on AI PCs are driving engineering apps like AutoCAD. “The idea is not to replace the architect,” he added.

Enterprises will take a hybrid approach to AI, with work being done on PCs, the edge, and in the cloud, Delgado said. “Everyone that tells you that everything is local is just trying to avoid the fact that the cloud models are improving at an exponential rate,” he said.

Most of the work doesn’t need cloud-level intelligence, and a lot of it can be done locally. Nvidia has tools that can redirect inferencing to idle PCs on a network — such as Macs and Windows PCs with Nvidia GPUs — to keep data in-house. It can also redirect to the cloud when needed.

“The next big thing in agents is inference routers that just point you to local hardware when it’s good enough, to cloud hardware when you need a bigger model,” Delgado said.

New hardware with fancy specifications is great, but it won’t deliver much if the AI engineer isn’t smart, said Deepak Seth, senior director analyst at Gartner. “You can give them the best tool, and they will still come up with the stupid stuff,” he said.

The solution to every problem is not just having more power, but “more power with the right people will lead to better results,” Seth said.

Even as they continue to release ever more capable competing models in a regular cadence, the top AI companies are joining forces to set AI safety standards.

According to The Information, Google, OpenAI, and Anthropic are reportedly working together to create a body tentatively called the Standards Authority for Frontier AI (SAFA). It would operate independently of government control, and set guidelines around risk assessment, testing, and pre-release review practices for frontier AI models. Sources close to the matter say the goal is to officially launch the initiative in early 2027.

The news comes in the same week as the heads of leading AI companies, including Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman, urged the United Nations to create safeguards around the very technology they’re building, to help prevent it from becoming too powerful to control. In addition, OpenAI this week posted a missive underscoring the importance of making continued AI progress “safe and beneficial.”

As Amodei and others warn of AI’s dangers, particularly when it comes to growing recursive self improvement (RSI) capabilities in models, enterprises want reassurance that they, and their customers, are safe from the growing perils of AI.

Ultimately, “enterprises care about AI in the same way they’ve cared about every other technology since the beginning of technology,” said independent technology analyst Carmi Levy. “The only real difference as AI blankets the technology landscape is the speed of change.”

Concerns around AI safety have ratcheted up in recent weeks and months, spurred by troubling incidents such as the attack on Hugging Face by autonomous OpenAI agents that broke out of sandboxes.

In its blog post this week, OpenAI contended that safe and beneficial AI progress requires advances in alignment research along with shared standards including agreed-upon baselines, common measurements, and incident reporting protocols. In fact, such international standards “may be as important to pacing the frontier as alignment research itself.”

Existing groups like the US Center for AI Standards and Innovation (CAISI), other federal frameworks, state regulations, and new forms of public-private partnerships⁠, can work together to advance this mission, the company said.

“Fully autonomous RSI is not happening today, and we should not pursue it unless and until it can be done safely,” OpenAI noted. “Whether and how to proceed must depend on our ability to preserve human control and on informed democratic choices about the benefits and risks.”

While regulation is indeed a growing focal point in the AI debate, the reality is that enterprises care less about the global state of AI safety than they do about how it will impact their day-to-day operational security, Levy noted. 

They’re concerned with whether an AI agent will expose corporate or employee data; platform vulnerabilities that could open the door to new forms of cyberattacks; hallucinations finding their way into “previously pristine” corporate workflows; and AI compromising compliance efforts.

“They’re worried about who’s responsible for what, when (not if) AI goes off the rails,” Levy said. They’re asking the same questions they have during previous technological disruptions like the cloud: What’s the ROI? What additional risks does AI impose? What new capabilities does it offer? Is the risk/reward worth it? Is the vendor responsive to our needs?

Enterprises also want assurance that the risks posed by emerging AI technologies, including agents, won’t increase corporate risk levels beyond their ability to monitor and respond.

But unfortunately, as the industry grapples with a summer’s worth of reports of agents breaking containment, roaming the internet, breaking into systems, and generally behaving in all sorts of unexpected ways, AI companies are in no position to provide these kinds of assurances, Levy said.

He advised enterprises to minimize their AI-centric risk profiles, and simultaneously maximize their ability to derive value from AI, by enforcing standards for vendors now, rather than waiting for something like SAFA to materialize.

This could involve requiring every new model to ship with the “equivalent of a safety and security datasheet” outlining, in detail, the model’s capabilities, known failure modes, and testing history. Extending existing change management processes to AI models is another means of minimizing deployment-related risk, Levy noted.

In addition, Yaz Palanichamy, senior advisory analyst at Info-Tech Research Group, pointed out, data privacy and IP leaks into public-facing AI frontier models are big concerns.

He advised enterprises to take great care to enforce effective acceptable use case policies (AUPs) around AI safety, and to consider establishing an internal AI proof of concept or governance committee. Equally crucially, they should obtain appropriate safety guarantees from their technology and AI vendors.

“Enterprises must proactively diagnose AI risk with the same level of seriousness as financial or cybersecurity risk,” Palanichamy said. This means creating a dedicated, cross-functional AI safety and ethics board committee composed of stakeholders from legal, cybersecurity, compliance, data engineering, and product development. 

“No AI tool should be deployed without the appropriate sign off,” he said.

The human element is important, as well: Enterprises should mandate AI literacy training to help users across the business understand where and how to spot hallucinations and requiring them to verify AI generated output prior to acting on it.

Other safety protocols to consider include real time toxic input filters and strict system prompt guardrails, Palanichamy advised. Also, methods such as continuous risk tiering can help categorize AI use cases by the level of risk they pose to the organization.

For example, a client-facing medical or financial analysis bot may require daily auditing and specific restrictions to ensure it does not infringe on sensitive information, “whereas an internal AI tool to help with summarizing public meeting transcripts may not need as much governance oversight,” he said.

This article originally appeared on CIO.com.

WordPress has patched what it described as a critical severity security vulnerability that would allow an unauthenticated attacker full remote code execution (RCE) capabilities. There have already been reports of attacks in the wild.

Given its popularity, WordPress has frequently been under attack, and patched another maximum severity bug allowing RCE in July. WordPress said the current hole, tracked as CVE-2026-87902, was discovered and reported to the company by Switzerland-based security researcher Robert Ressl.

The post announcing the WordPress 7.1.2 security release said that the fix addressed an issue where “an unauthenticated attacker can, under certain conditions, make page template resolution include a chosen readable local PHP file outside the active theme directories. If relevant pre-conditions for both the server environment and the active theme are met, this can lead to remote code execution (RCE).”

It urged users to update their sites immediately, and said that the fix had been backported through version 4.7, as the hole also impacts many older versions of WordPress.

Noah Kenney, principal consultant at Digital 520, said his top concern was the massive extent of damage an attacker could inflict with this hole. 

“Once attackers have PHP execution, they can read wp-config.php, obtain database credentials and authentication keys, create administrator accounts, alter payment or lead capture forms, redirect visitors, and install persistent code,” he said. “The less obvious part is how they get there. Pearcmd.php is a legitimate PHP package management tool, but attackers can abuse its configuration commands to write content of their choosing to disk. The current attacks use it to place malicious PHP in /tmp, then use the WordPress flaw to load and execute that file. A security team monitoring only the WordPress directory for changes could miss that first stage entirely.”

But IDC and others saw the speed with which attacks began as the most concerning element of this report. 

“This WordPress vulnerability is a textbook example of the new risk reality: attackers are exploiting critical bugs within hours of disclosure, and most enterprises are not patching fast enough to keep up,” said IDC Research Director Philip Harris, citing reports from security firm Patchstack that attackers had leveraged the hole shortly after WordPress posted the patch. 

Patchstack said, “When this post first went up, every request we had seen was reconnaissance against harmless core files. That is no longer true. Attackers are now including pearcmd.php and using it to write PHP files to disk and public scanning tooling for this CVE is in circulation.”

IDC’s Harris noted that Patchstack’s telemetry illustrated the new reality, with attacker reconnaissance occurring within five hours of the patch, and full exploitation within about a day. “The window between disclosure and exploitation has collapsed to the point that mean time to exploit for critical vulnerabilities is now negative in some cases, meaning exploit code appears before or immediately after a patch ships,” he said. “This WordPress bug tracks that pattern closely: Patchstack recorded the first probing traffic under five hours after WordPress 7.1.2 was released, and traffic volume increased roughly tenfold within a day as attackers moved from scanning to actual payload delivery.”

Aman Mahapatra, chief strategy officer for New York City-based technology consulting firm Tribeca Softech, agreed.

“The number that should anchor this story is not the 9.2 CVSS score, but it is the gap between patch and exploit. With this vulnerability, that gap was effectively zero,” he said. “WordPress shipped 7.1.2 on September 22, and Patchstack blocked the first exploitation attempt at 11:49 UTC the same day, using payloads that matched the exact encoding the patch was written to fix. Attackers did not discover this bug. They read the fix. Publishing a patch is now functionally publishing an exploit guide, and any enterprise still running a remediation cycle measured in weeks is operating on a timeline that stopped existing some time ago.”

One response to the speed of attacker actions is to automate updates, but that is not necessarily a good option. Some enterprise CISOs are nervous about excessive patch automation; they want to review and approve any system changes, mostly to avoid disasters such as 2024’s Crowdstrike incident. 

Ressl said in an interview that another concern about automated updates is that many enterprises don’t perform sufficient verification that the updates executed properly.

“Having automatic updates enabled is not the same as verifying that the patch is installed,” he said. “Compatibility and availability concerns are legitimate reasons for testing updates. My recommendation is a rapid, tested rollout with verification across all exposed installations, including staging sites.”

Mahapatra added that the problem with update automation is especially severe in larger enterprises. 

“WordPress applies minor security releases automatically by default, which means the average hobbyist blog is probably already patched,” he pointed out. “Enterprise sites routinely disable those auto-updates to enforce change control, so the organizations with the most mature governance are the ones most likely to still be exposed this week, because their own process is holding the fix in a queue while attackers scan. Change control that cannot distinguish an unauthenticated remote code execution flaw from a routine plugin update is protecting the process rather than the enterprise, and this is the vulnerability that makes that distinction expensive.”

And, noted Nidhi Luthra, executive advisor for Acceligence: “A security release can become an attacker roadmap almost immediately, so emergency patching for internet-facing systems needs an hours-level response.”

Mahapatra added that another, perhaps bigger, concern is the fact that many enterprise WordPress deployments happen below the radar. They are not typical shadow IT, because they are fully authorized at the time, but they are still often unknown to IT management. 

In enterprises, he said, “the exposure is materially larger than most organizations will assume because most enterprises do not think of themselves as WordPress shops, and almost all of them are. The risk sits in the web estate that IT never inventoried: marketing microsites, campaign landing pages, regional country sites, investor relations pages built by an outside agency, and the properties of companies acquired three years ago that nobody got around to migrating. The affected [WordPress version] range runs from 4.7.0 through 7.1.1, nearly a decade of installs, and the forgotten sites are precisely the ones still sitting on old branches.”

Mahapatra said that he has especially seen this in enterprises in the finance vertical.

“When I walk through external attack surface reviews with banking CISOs, the WordPress instances that surface are almost never the corporate site,” he said. “They are the sites that belong to marketing, a subsidiary, or an agency contract that expired, and none of them appear in the CMDB the security team is patching against.”

This article originally appeared on CSOonline.

Microsoft 365 E5 and E7 customers can now run security information and event management (SIEM) inside Microsoft Defender at no extra license cost.

Microsoft is delivering the capability through the Integrated Security Operations Center (ISOC) in Microsoft Defender, which combines SIEM with Defender’s existing XDR, threat intelligence, automation and AI tools in a single portal.

“Security cannot operate at AI speed when protection and operations are built as separate systems,” Rob Lefferts, corporate vice president of Microsoft Threat Protection, wrote in a blog post announcing ISOC. He said attackers now use AI agents to automate attacks at scale, and that every handoff between separate tools slows defenders down.

Until now, E5 and E7 included Defender XDR but not SIEM, which Microsoft sells separately as Microsoft Sentinel.

Under ISOC, logs from Microsoft’s own security products are available without ingestion charges. From Oct. 1, data ingested from third-party security tools and other outside sources will be metered at $2.40 per GB on a pay-as-you-go basis, Microsoft said in a technical blog post.

The company opened ISOC as a public preview, meaning it is not fully production-ready, on Sept. 23 and has not said when the preview phase will end. It is available to organizations with an active Microsoft Defender Suite and Microsoft 365 E5 or E7 license that do not already run Microsoft Sentinel, and has no minimum seat requirement.

For organizations whose security stack already runs largely on Microsoft, the trade is attractive, said Yih Khai Wong, senior research manager for security services research at IDC Asia/Pacific.

“Most telemetry already resides in Defender, Entra, and M365, so bundling Sentinel’s SIEM into E5/E7 effectively makes ingestion free, with no separate license needed to move and re-analyze data your own vendor already holds,” Wong said.

“ISOC is not a new standalone product,” Microsoft said in an FAQ accompanying the technical post. Its ISOC product page said capabilities such as workbooks and natural-language-to-SOAR automation “previously required a separate Sentinel purchase.”

According to the technical post, Microsoft began rolling out case management, workbooks and natural-language playbook generation to eligible customers’ Defender portals on Sept. 23, with no additional configuration required.

The included data covers Defender for Endpoint, Office 365, Identity, Cloud Apps and Cloud, plus Microsoft Entra ID Protection logs and Azure and Office 365 activity logs. Retention is 30 days during the preview, rising to 90 days on Nov. 15, the post added.

Anything beyond that requires an ISOC workspace, which needs an Azure subscription, according to Microsoft’s product documentation. The workspace unlocks more than 500 data connectors, user and entity behavior analytics (UEBA), CI/CD repositories and threat intelligence.

Wong said mixed environments face a different calculation. “For multicloud shops or organizations with significant non-Microsoft telemetry, this requires a real total cost-of-ownership comparison against current SIEM spend.”

The current preview excludes organizations with an active Sentinel workspace. Microsoft said nothing changes for existing Sentinel customers, and those meeting the licensing criteria can choose to move to ISOC from Nov. 15.

Consolidating identity, endpoint, productivity and security operations under one vendor can simplify investigations, Wong said, but it also “creates greater dependency on that vendor’s roadmap, pricing model, availability, and security resilience.”

CISOs running a separate SIEM face a genuine evaluation, not a quick swap, weighing migration effort, retraining of detection content, exit costs and contractual complexity, he added.

Sanchit Vir Gogia, chief analyst at Greyhound Research, advised enterprises interested in ISOC to “reproduce the organisation’s highest-value cross-vendor detections before replacing a mature SIEM,” warning that “Microsoft’s claim of more than 500 connectors establishes reach, not equal treatment of every source.”

Microsoft’s Lefferts positioned ISOC as the foundation for the agent strategy Microsoft introduced alongside Project Perception in July.

Gogia questioned the premise that this requires a single vendor. “Microsoft has established a useful integration path. It has not established that one supplier is technically necessary for an agentic SOC,” he said. He noted that Project Perception remains invitation-only in limited public preview.

Microsoft has not published details on how agent actions within ISOC are logged, audited or reversed. Gogia said the critical threshold comes when agents gain authority to make changes, since changes to identity, endpoint or cloud policy “create a blast radius.” He recommended distinct agent identities, narrow tool permissions and independent approval for consequential changes.

The agents themselves add attack surface, Wong warned. “Adversaries may attempt to manipulate the telemetry an agent relies on, poison contextual data, or use prompt injection techniques to influence agent behavior and decision-making,” Wong said.

“Microsoft has made integration easier to buy. The enduring test is whether customers retain the means to challenge and reverse it,” Gogia said.

Microsoft did not immediately respond to a request for comment.

This article first appeared on CSO.

Jamf was a pioneer in Apple device management when it began in 2002. The company was ahead of its time: its founders could see that Apple had a future in the enterprise, but not many others saw it. Jamf now runs 35.2 million devices from 78,000 organizations. And it’s still looking to the future.

I spoke with Jamf CEO Beth Tschida, who joined me for a chat directly after her keynote speech at JNUC, Jamf’s big event for Mac admins, on September 23. 

Tschida wasn’t there to rehash Jamf’s past for relevance in the present, but to share the company’s deep commitment to being where its customers are going to go. That means artificial intelligence, which in Jamf’s case involves weaving AI across its ecosystem of products, from setup to tech support, security, and beyond.

It’s not a new vision — you could argue that some early signs of its intention were visible when the company acquired ZecOps in 2022 — but time moves forward and the company has coalesced around a firm philosophy of how to use AI, where to use AI, and how AI should be deployed in a human-centric way for the benefit of its customers.

“AI runs better on Apple, and Apple runs better on Jamf,” Tschida told me. The next piece to that statement is to consider how AI can be effectively managed by IT, she added. “You have to be able to see it. You have to be able to govern it, and then you want to be able to harness it.”

This points directly to shadow AI, one of the big-picture problems IT has with AI at the moment as employees feed confidential data to the cloud-based AI models. Jamf’s response has been to create new frameworks IT can use to monitor use of AI across their managed fleets. These tools let IT identify use, manage use, or even stop use altogether.

“How can we look at that, see it, govern it, and harness it in the ways that AI is providing that opportunity, right at the endpoint?” Tschida said. 

The other component is pricing. Even when data is legitimately shared to AI, the cost of inferencing is high and getting higher. Jamf now offers tools to let IT monitor this use and the cost of it, enabling management to suggest lower cost or on-prem AI solutions for tasks where appropriate. After all, why would you pay for Fable when you can get your Apple device to help write that letter?

Tschida explained that part of Jamf’s response to AI costs is the inclusion of granular model controls, which let admins assign different teams access to different models, the idea being to prevent what she called “overmodeling” — the use of expensive high-reasoning models for tasks that lightweight models can handle.

“You have to understand what the need is. You must lay it out there in a governed way. Human in the loop and then you harness it,” she said.

But the Jamf vision doesn’t end with AI management; it extends to making active use of the tech. In this case the company has introduced AI-powered tech support for routine problems to free up staff time. There are two strands to this approach: More obviously, easy to access tech support should reduce time spent on routine problem solving, but another compelling aspect of this evolution is the move toward proactive device health.

Traditionally, IT has been reactive: a user experiences slowness, they eventually file a ticket, and IT fixes it. But many users simply “endure” suboptimal performance without ever reporting it, Tschida explained. “A lot of times, users don’t even raise a ticket. They just endure.”

That was then, but the future looks different. “We believe that we now have the right telemetry to look at device health,” Tschida said. “You’re really moving ahead of problems. Users don’t even raise a ticket… you can help them optimize it without them knowing it.”

With over 35 million devices now managed using Jamf, the origin story of the company seems further ahead of its time now than ever. Apple’s enterprise product marketing lead, Jeremy Butcher, appeared at one point during Tschida’s keynote to talk about where Apple now is in the enterprise. He pointed to data that Macs generate 55% fewer help tickets than Windows systems. He confirmed that the MacBook Neo has accelerated enterprise growth for the platform, and pointed out that more Apple devices are purchased in enterprise than any other brand.

With so much energy behind Apple’s platforms, what has Tschida been seeing?

“We know that Anthropic and other companies, they tend to deploy for Mac first. The developers are mostly running Macs. That’s what they prefer to be on, and they’re at the forefront of agentic AI.”

The thing about the growing enterprise Apple ecosystem is that no single company can build the perfect set of solutions for every enterprise. Part of Jamf’s response to this has been the creation of platform APIs, which customers and partners can use to extend the platform to meet their needs. “I love a good API because it allows you to run a road map way faster,” Tschida said.

Former Jamf CEO Dean Hager used to say, “When Apple innovates, Jamf celebrates.” 

His successor echoes his point, with a focus on what Jamf can do for its customers. “Our job is to make sure that Apple runs better with Jamf,” she said. “We’re scaling it, we’re securing it, and we’re making sure that you can… first see it, and govern it, and then harness what you can get from it in a way that’s helpful for our customers.”

A second aspect to governance is around sovereign, private, and/or on-premises AI. I asked Tschida what her customers are saying. “There will be some that look at the benefit if you can run at least some of your workflows locally,” she said. “There’s a tokenomics piece of it. There’s a privacy piece of it. There’s a sovereignty piece of it.”

Whatever the motivation, for Jamf the question remains, “How can you determine what we can do and help you do, locally on device, and build that so you can run it at scale? That’s where we’re focused.”

I walked away from the interview feeling a vision for Apple in the enterprise in which IT is augmented by smart tech infrastructure: Macs that request support before users know they need it, systems that self-identify bottlenecks or wasted resources such as over-generous token consumption and let IT take steps. A more intelligent enterprise infrastructure, equipped with real-time awareness, self-reliant with little to no cloud exposure.

For Apple, Jamf, and certainly for the Mac, that seems like several giant steps since the foundation of Jamf in 2002. AI in IT is an accelerant, and Jamf is working with it. “We are doing our best to run at the speed of AI,” said Tschida. 

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

It’s been five years since Google first offered a glimpse of Project Starline, a three-dimensional videoconferencing platform billed at the time as as a “magic window” that enables realistic and immersive calls.

On Wednesday, Google announced that Google Beam — as the software element of the product is now called — is generally available for purchase, built into HP’s Dimension video meeting hardware. The Dimension hardware reportedly costs around $25,000 per unit; Google did not provide details for costs related to hardware or additional software licenses. The combined products will be sold in six countries at first: Canada, France, Germany, Japan, UK, and US.

Google has also partnered with co-working space provider Industrious to make Beam available to try at its locations across the US, including Atlanta, Chicago, New York City, and Palo Alto.

Beam “offers the potential for immersive, 3D engagement, without the challenge of wearables,” said Irwin Lazar, principal analyst at Metrigy.

“This is one of those ‘you have to try it’ experiences to understand the value proposition,” said Avi Greengart, president and founder of Techsponential. “It really does feel like you’re talking to someone a half a world away, with all your micro-expressions and nonverbal cues intact.

“This doesn’t just save you travel time and budget, it enables closer collaboration throughout a project or working relationship that provides different value than occasional live visits, which you may still need.”

Google began developing the Beam technology several years before it was unveiled amidst the COVID-19 pandemic and shift to remote working as offices were shuttered. It combines AI-generated 3D images of participants with spatial audio to create the sense of being in the same room.

The hardware has evolved considerably since then, with Google and HP shrinking the original large video booth into a smaller system that more closely resembles a conventional videoconferencing setup.

The 65-inch Dimension display renders meeting participants in their full size, and includes the cameras, 12-microphone array, and LED lighting required for the Beam software. Google Meet and Zoom video meeting apps are supported.

While Beam currently enables one-to-one meetings, Google recently highlighted its efforts to enable group calls, too. This is currently at an experimental stage.

Google claimed that an internal study showed 50% of Google staffers who used Beam felt more connected, resulting in a 21% drop in the need for follow-up meetings.

Select customers were granted early access last year, ahead of today’s full launch. One of them, management consulting firm Bain, uses Beam to interview job candidates remotely, reporting that it helps interviewers pick up on nuances of body language and eye contact without needing to meet in person. This also saved on travel costs, the company said in a promotional video.

It remains to be seen whether the technology will gain a wider audience. The price tag will “present somewhat of a barrier to adoption,” said Lazar. He expects early adopters will target specific use cases such as “executive meetings, hiring interviews, product demos, and high-stakes engagement.”

Greentgart described Beam as a “network-effect system,” as the technology is required on both ends of a Beam call. “That makes the expansion of Google Beam on HP Dimension to more locations crucial,” he said. “I expect that some companies will try it at a conference center before investing in endpoints inside their own companies.”