Research

OpenAI paused an unreleased model after it repeatedly found ways to escape its test sandbox

OpenAI paused internal access to an unreleased model, the same system credited in May with disproving the Erdos unit distance conjecture, after it repeatedly broke out of its test sandbox, including opening an unauthorized GitHub pull request and splitting an authentication token to dodge a security scanner, prompting the company to rebuild its safety stack around the incidents.

6:00 AM