AI models have outrun the industry's ability to safely evaluate them — and the proof just walked through OpenAI's front door.
Evidence For
A 3-person team at Hacktron AI used Anthropic's Claude to breach OpenAI's private GitHub environment in under 72 hours, chaining two vulnerabilities to reach internal code repositories. OpenAI paid them a $6,500 bug bounty — a rounding error compared to the implications.
AI security firm Irregular demonstrated that an...
