OpenAI has admitted that ChatGPT can go rogue and delete people’s files without telling them.
In recent days, numerous users of Codex – the programming platform built into the AI assistant – reported that the tool appeared to have deleted their files without being asked to and without getting permission.
One engineer, Bruno Lemos, said that it had “deleted my whole production database”. Another AI investor called Matt Shumer said it had “accidentally deleted almost ALL of my Mac’s files”.
“Multiple such incidents have been reported,” said Gary Marcus, an artificial intelligence expert who has repeatedly warned about the dangers of current products.
“A clear reminder that current AI cannot be trusted. In racing these techniques ahead, we are asking for trouble”
The problem appeared to be a result of a new version of the model that powers Codex, known as GPT-5.6 Sol. Before the release of the update, OpenAI had warned that the system was liable to make potentially dangerous decisions on its own.
Now OpenAI has confirmed that the behaviour is happening. It appears to happen “most commonly” when users give the system control over their computer and does not have protections enabled, which then leads it to make “an honest mistake and mistakenly” and delete people’s files, said Thibault Sottiaux, one of OpenAI’s product leaders.
“This is of course not how we want the system to behave, even when a user operates the model in full-access mode without the safeguards of our sandbox or without using auto review which checks for these kinds of high risk actions and rejects them,” he wrote on X, formerly known as Twitter.
He said that OpenAI would take steps to limit the behaviour including showing a different message to developers, advising them to turn on protections, and adding safeguards to ChatGPT. “Even though this happens extremely rarely, we’ll share a detailed post-mortem in the coming days that goes into more details and what we are doing to minimise risks further,” he wrote.
