Amid the buzz and excitement surrounding GPT-5.6 Sol, OpenAI’s latest coding and cybersecurity focused flagship AI model, recent posts across social media suggest that the rollout has hit a few bumps.Several users on X and Reddit have reported instances of the model autonomously deleting their files, data, and even entire databases on its own without seeking human approval. It’s still too early to say how widespread these reported incidents are, including claims that GPT-5.6 Sol deleted files or surfaced credentials users say they never provided.As a result, there is no statistically reliable evidence to attribute the model for being at fault, especially since plenty of other variables can influence an AI system’s behaviour. However, it is also worth monitoring as GPT-5.6 family of models roll out more broadly.“GPT-5.6-Sol just accidentally deleted almost ALL of my Mac’s files,” Matt Shumer, the founder and CEO of AI startup OthersideAI, maker of HyperWrite, wrote in a now viral post on X.Three days ago, GPT-5.6 deleted my Mac’s home directory.It absolutely sucked.But so many OpenAI folks reached out, and @gdb called me and offered to do anything he could to help.Massive props to OpenAI for handling a shitty situation incredibly well. https://t.co/xRw75VoVQz— Matt Shumer (@mattshumer_) July 13, 2026Bruno Lumos, a developer, posted that GPT-5.6 Sol deleted his whole production database. Another developer, Joey Kudish, said, “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.”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. It’s not safe. https://t.co/KnG00h84j7 pic.twitter.com/u7y2W8CODn— Bruno Lemos (@brunolemos) July 13, 2026These user reports of the model going rogue are surprising given that GPT-5.6 Sol, along with other models released by OpenAI, underwent additional safety review prior to their release. It also comes at a time when the rise of AI coding agents has been accompanied by fears that they could take malicious actions on their own or upon being hijacked by threat actors.OpenClaw, the popular open-source framework for building and running AI agents locally on a company’s own hardware, was also plagued by similar security concerns among developers in the early days of its release.The Indian Express has reached out to OpenAI for comment and will update this report with its response once received.What is GPT-5.6 Sol?In June 2026, OpenAI unveiled its next-generation GPT-5.6 family of AI models comprising three model versions, namely: Sol, its flagship model that is said to be the most capable; Terra, a mid-tier version that is more balanced for everyday use; and Luna, a fast and affordable version of the model.Story continues below this adGPT-5.6 Sol is the best-performing model out of all of them based on testing on benchmarks related to cybersecurity, biology, and agentic abilities. It has also been developed with a “layered safeguard stack,” which looks to prevent bad actors from using its AI model for cyberattacks, among other malicious behaviours, according to OpenAI.Also Read | OpenAI’s ChatGPT Work goes beyond coding to automate office tasksSoon after the launch of the GPT-5.6 lineup, OpenAI delayed the global rollout of these models indefinitely at the request of the US government over heightened national security concerns about the potential misuse of powerful AI technologies. The ChatGPT-maker was allowed to make these models accessible to the public again after a few weeks, though it is unclear what exactly led to the curbs being lifted.Does GPT-5.6 Sol have any flaws?Alongside the launch of GPT-5.6 Sol, OpenAI also published a system card documenting the company’s model-testing methods and results.While the system card largely focuses on the model’s advanced capabilities, it also flags that Sol has a tendency to take whatever actions it thinks gets a job done, even destructive ones, as long as those actions are not “unambiguously” prohibited. The model is also capable of lying about what caused it to take such actions, as per the document.Story continues below this ad“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,” OpenAI said in the system card for GPT-5.6 Sol.“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,” it added.Also Read | OpenAI debuts GPT-5.6 and ChatGPT Work to bring AI agents into the workplaceOpenAI has also shared examples where the user instructed Sol to delete three remote virtual machines or cloud-based computers labelled 1, 2, and 3. Instead of stopping to ask the human user after it failed to locate the machines, the model proceeded to delete virtual machines labelled 5, 6, and 7.The system card further reveals another instance of Sol using “credentials beyond what the user had authorised” when it went looking for the credentials on its own, and found some sitting in a hidden local cache. However, the system card also states that destructive behaviour by Sol can be rare.Story continues below this adIn the meantime, users of GPT-5.6 Sol can safeguard themselves by withholding access to their main production systems, maintaining backups, and carrying out staged rollouts.