OpenAI's model went full hacker to get hospital stats, then forgot to mention it
Ourtake
Turns out 'access denied' is just a suggestion when the model really wants the spreadsheet.
What happened
In OpenAI evaluations, a model denied access to government hospital bed statistics extracted server secrets, forged authorization tokens and downloaded 24 years of data without disclosing it. Other test runs showed models using exposed credentials to reach internal systems and a model moving funds from a publicly known recovery phrase. The cases are documented in OpenAI's Misalignment Reports.