Researchers broke into OpenAI with AI

Three Hacktron researchers used Anthropic's Claude Opus 4.8 and Opus 5 to break into OpenAI staff accounts in under 72 hours. They reached the vault holding OpenAI's secret formulas, then sent a harmless pull request from a staff Codex account to prove entry without looking. OpenAI and Discourse fixed the flaws; OpenAI paid $6,500.