OpenAI’s Codex Was Accidentally Nuking User Files Like a Digital Janitor
It turns out giving an AI free reign over your terminal is a recipe for digital chaos. OpenAI is frantically patching Codex after the bot started treating important user directories as glorified trash bins.
When Codex was tasked with cleaning up temporary files, it occasionally decided that the entire Windows user profile was just a particularly large pile of digital garbage. The issue stemmed from a collision between the PowerShell variable $home and the system-level $HOME, causing the AI to target the wrong directory with extreme prejudice. It wasn't just a simple mistake; the agent was essentially playing Russian roulette with file paths.
In another spectacular display of incompetence, the model ignored bind mount restrictions and started wiping data inside actual repositories instead of temporary folders. OpenAI has now implemented strict new guardrails: the AI must perform a identity check on the target folder before executing any command, and it is explicitly forbidden from touching paths like ~ or $HOME. Dangerous deletion commands now require a secondary verification layer, forcing the model to retreat to a safer path if it can't prove its target is truly expendable.
The company is now shoving users toward sandboxed environments, admitting that giving an AI Full access is basically an invitation to system suicide. While the developers are busy patting themselves on the back for adding more training simulations to catch these errors, the reality is that the gap between a chatbot’s "logic" and the messy reality of OS environment variables remains a canyon-sized liability. It’s almost impressive how quickly a sophisticated neural network can be turned into a destructive script kiddie with the right set of permissions.
Source: Twitter
Comments
Help shape the next version: Add context or suggest a correction. AI review can add points toward a rewrite. Reviews and updates may take time; a full meter does not guarantee a new version.