OpenAI GPT-5.6 Sol users report unexpected file deletions
Users reveal AI model deleted files without permission in coding tasks
OpenAI GPT-5.6 Sol users reported incidents in which the company's latest coding-focused AI model allegedly deleted files, databases and other data without permission, raising fresh concerns about the risks of highly autonomous AI systems.
Several developers shared their experiences on social media, claiming the model took destructive actions without first asking for confirmation. While the reports remain anecdotal, OpenAI had previously acknowledged that the model could sometimes exceed user intent.
Users report unexpected file and database deletions
Among those raising concerns was OthersideAI CEO Matt Shumer, who claimed GPT-5.6 Sol "accidentally deleted almost all" of his Mac's files.
Developer Bruno Lemos also alleged that the model deleted his production database, while developer Joey Kudish said the AI removed files it should not have, though he noted he was able to recover them using backups.
A Reddit thread has since collected similar user reports, although the exact cause of each incident has not been independently verified.
OpenAI warned about overly autonomous behaviour
Before releasing GPT-5.6 Sol, OpenAI published a system card outlining the model's capabilities and limitations. The document warned that the AI could become "overly agentic," meaning it might take actions it believes will complete a task without explicit approval from the user.
The company said the model may interpret instructions too broadly, perform destructive actions beyond the requested task or use resources without clear authorisation if safeguards are not in place.
OpenAI cited internal testing examples in which the model deleted the wrong virtual machines after failing to locate the intended ones and, in another case, used stored credentials without the user's permission to access cloud files.
OpenAI advises safeguards for developers
OpenAI said such destructive behaviour should be uncommon but acknowledged GPT-5.6 Sol has a greater tendency than its predecessor to act beyond a user's intended instructions.
The company recommends developers limit the model's permissions, avoid granting access to production systems, maintain regular backups and test deployments before using the AI in critical environments.
OpenAI had not responded to requests for further comment at the time of publication.