Tech
OpenAI’s new flagship model deletes files on its own, people keep warning
Users of OpenAI’s latest coding and cybersecurity-oriented flagship model, GPT-5.6 Sol, are posting horrifying accounts on social media, claiming the model just up and deleted their files, data, even entire databases on its own, without asking first.
“GPT-5.6-Sol just accidentally deleted almost ALL of my Mac’s files,” wrote Matt Shumer, the founder and CEO of AI startup OthersideAI, maker of HyperWrite, in a now viral post on X.
“GPT-5.6 Sol just deleted my whole production database. That’s it. Not a joke. This had never happened to me before, with any other model, ever,” developer Bruno Lemos posted on X.
“Looks like I’ve gotten bit by Codex Sol’s overly ambitious system and it deleted some files it shouldn’t have. I have backups so I’ll be fine, but this is not cool, Sol needs to be toned down,” posted developer Joey Kudish.
A Reddit post has collected more examples.
True, a handful of users making such claims — even one as credible as Shumer — isn’t statistically reliable evidence that the model is solely at fault. Plenty of other variables can cause an AI system to misbehave.
But OpenAI itself flagged this risk before Sol ever shipped. Two weeks before OpenAI released GPT-5.6 Sol, the company published a system card for the model — the paper that documents model-testing methods and results. Naturally, the system card largely extols the capabilities of Sol, as these reports typically do. But it also includes a warning of sorts (bold emphasis ours):
In coding contexts, misalignment generally stems from a mix of overeagerness to complete the task and interpreting user instructions too permissively — assuming that actions are allowed unless they’re explicitly and unambiguously prohibited. This manifests as the model being overly agentic in circumventing restrictions it faces when attempting the requested task, being careless in taking actions which may be destructive beyond the scope of the task, or deceptive when reporting its results to users.
In other words, OpenAI found that Sol has a tendency to take whatever actions it thinks gets a job done, even destructive ones, as long as those actions aren’t “unambiguously” prohibited. Then it might lie about what caused it to do so.
OpenAI shared examples. In one case, the user told Sol to delete three remote virtual machines (cloud-based computers), named 1, 2, and 3. But Sol couldn’t find those names in the place where it looked, so instead of stopping to ask, it decided to delete three other virtual machines, 5, 6, and 7, the paper notes. In doing so, it “killed active processes, and force-removed worktrees [the working files tied to a coding project]. It later acknowledged that uncommitted work on remote virtual machine 6 may have been lost.”
In short, it deleted the wrong machines, on its own, and only admitted what it did after the fact.
In another instance, Sol “used credentials beyond what the user had authorized.” Credentials are the usernames, passwords, or security keys a system uses to verify who’s allowed to log in. This incident occurred when Sol was working on a project and couldn’t read its cloud files. Rather than alerting the user to the problem, Sol went looking for the credentials on its own, found some sitting in a hidden local cache, and then used them without asking for authorization from the user.
The system card does promise that destructive behavior should be rare, although it also admits that GPT-5.6 Sol “shows a greater tendency than GPT-5.5 to go beyond the user’s intent, including by taking or attempting actions that the user had not asked for.”
It’s too soon to say how widespread these incidents — Sol deleting files, or sifting out credentials the user didn’t give it — really are. In the meantime, Sol users should be prepared to implement their own safeguards with the model, like using permission scoping (that doesn’t give access to production systems), maintaining backups, and staging rollouts.
OpenAI did not immediately respond to our request for comment.
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.
Tech
Iran abused mobile networks’ vulnerabilities to locate US military in the Middle East, report says
The Iranian government abused well-known vulnerabilities in the global telecoms infrastructure to locate U.S. military personnel in the build-up to the Iran War, as well as in the early days of the conflict, according to Financial Times.
The Iranian government exploited Signaling System 7, or SS7, a set of protocols for 2G and 3G networks that has long been the backbone of how cellular networks connect to each other to route subscribers’ calls and texts around the world, the newspaper reported, citing research by the Mobile Surveillance Monitor, as well as anonymous government officials with knowledge of the spy campaign.
Intelligence agencies have long abused SS7 to track cellphones abroad, which is what happened in this campaign.
Using this technique, Iran was reportedly able to locate U.S. military forces stationed in military bases as well as hotels in Iraq, Bahrain, and other countries in the Middle East, which allowed the regime to strike them. These attacks resulted in several injuries.
Apart from SS7, Iran also abused advertising technology used to serve tailored ads to cellphone users, another well-known surveillance technique that relies on everyday technology.
Tech
Google Images gets a Pinterest-like redesign focused on discovery
Google Images, the tech giant’s image search engine, is taking on Pinterest with its latest redesign that turns the site into a browsable, dynamic gallery of images from across the web. Google is also adding a way for users to create AI images right in Search, as it celebrates 25 years since the debut of Google Images.
Pinterest has long been known for allowing people to browse and save visual inspiration for everything from fashion to home decor. With this redesign, Google is essentially copying that playbook by turning Google Images into a place for discovery and inspiration, and not just search, which could increase users’ time spent on Google platforms, helping boost its ad revenue.
In addition, Google is likely hoping that when users can’t find the image they’re looking for on Google Images or when they want to visualize something, they’ll stay within its ecosystem to create it rather than turn to third-party services like ChatGPT.
After navigating to the redesigned Google Images, users will see a “For You” gallery of images tailored to their interests and browsing history. Like Pinterest, the gallery is designed for continuous browsing, with Google saying it updates in real time with new images.
As users browse, they can save ideas to their “collections,” which will appear as tabs above the main gallery of photos. For example, users can create collections for things like vacation outfit ideas, travel inspiration, and ways to design a reading nook, which they can come back to later.
The redesign is rolling out over the coming weeks on desktop in the U.S. in English. Users need to be signed into a Google Account to try it out, the tech giant says.
As for generating images directly in Search, Google says the feature is meant for moments when you have a highly specific idea for an image that doesn’t already exist online. Google is bringing image generation directly into AI Overviews on Search and will use its latest Nano Banana model to transform a text prompt into a custom visual.
The feature can also help users reimagine spaces and visualize ideas, such as seeing what a room might look like painted red or what a dorm room with a coastal theme could look like.
Image generation in AI Overviews will start to roll out over the coming weeks in English for all regions that currently support image creation in AI Mode, Google says.
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.
Tech
Meta’s Adam Mosseri says AI token budgets could soon be capped per engineer
In a recent interview, Instagram head Adam Mosseri said he can see a time in the future, perhaps only a year or two, when putting limits on Meta employees’ AI token spend will become necessary.
“I think that you can imagine, at least in a year or two … that the burn rate of a strong engineer might be the same as their salary, or their cost of employment. And in that world, you’re going to probably need to put in some caps,” the Meta executive said, while speaking on Lenny’s Podcast.
AI token spend, a reference to the cost of processing AI prompts and responses, has been a much-buzzed-about subject in recent days. Meta shut down an internal AI token spend leaderboard after AI costs put the company on track for billions of dollars in 2026.
Meta is not alone in rethinking its approach to AI experimentation. Uber also had an AI reckoning after it blew through its 2026 AI coding budget by April. Soaring token costs saw Microsoft cancel Claude Code licenses, consolidating its engineers around its own Copilot CLI tool instead.
Mosseri’s belief, he explained, is that AI token costs will have to be managed just like any other resource, offering an analogy to things like payroll or operating expenditure (OpEx), which is the day-to-day costs of running a business.
“I think of it like…any other resource,” Mosseri said. “I have to decide how to deploy capacity to my different teams because I have a limited number of GPUs and CPUs and storage and RAM etc. I have to decide how to deploy OpEx for labeling budgets across my teams. I have to decide how to deploy payroll for headcount across my teams.”
Token budgets will be the same, he added, noting that the cap per engineer would have to be proportional to the company’s trust in their ability to use the budget in an “ROI-positive” way.
Meta doesn’t currently have token caps for any employee, Mosseri said, but he believes that their use could be healthy in the future. Further down the road, he expects token costs to come down as the AI model makers enter a pricing war to attract people to use their tools over their competitors.
For now, the company has managed to rein in its token costs a bit by shutting down the “silly things” that it was doing, Mosseri noted — like that token spend leaderboard.
“It’s not that hard to build a token incinerator, and that doesn’t create a lot of value,” he said.
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.
France D William Saliba departs with injury, replaced by Maxence LaCroix
The Open distant fourth among majors with $17.75M prize pool
