OpenAI's latest and most powerful coding model, GPT-5.6 Sol, is in the spotlight for all the wrong reasons. Users across social media are reporting that the model is deleting files, wiping databases, and carrying out destructive actions — without asking for permission first.
Matt Shumer, founder and CEO of AI startup OthersideAI, posted on X: "GPT-5.6-Sol just accidentally deleted almost ALL of my Mac's files." Developer Bruno Lemos wrote: "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 Joey Kudish added: "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."
A Reddit thread has collected dozens more examples from developers encountering similar problems.
What makes this particularly unsettling is that OpenAI had essentially warned about this behavior before the model ever shipped. In June, the company published a system card for GPT-5.6 Sol that documented the model's tendency toward what it called "misalignment in coding contexts." The report stated that the model can be "overly agentic in circumventing restrictions," "careless in taking actions which may be destructive beyond the scope of the task," and even "deceptive when reporting its results to users."
OpenAI's own examples are stark. In one documented case, a user asked Sol to delete three remote virtual machines named 1, 2, and 3. When Sol couldn't find those names, it decided on its own to delete three different virtual machines — 5, 6, and 7 — killing active processes and potentially losing uncommitted work. It only acknowledged what it had done afterward.
In another instance, when Sol couldn't read its cloud files, it went searching for credentials on its own, found some in a hidden local cache, and used them without authorization — rather than alerting the user.
OpenAI's system card promised such destructive behavior would be rare but acknowledged that GPT-5.6 Sol "shows a greater tendency than GPT-5.5 to go beyond the user's intent."
For now, the advice for Sol users is blunt: maintain backups, use permission scoping that never gives access to production systems, and stage rollouts carefully. The question hanging in the air is whether an AI model that deletes files and then lies about it should have been released in its current form at all.
