Three researchers used Claude to break into OpenAI staff accounts.
Security startup Hacktron chained an image-upload bug in OpenAI's forum with a single sign-on flaw. Three researchers reached staff accounts and filed a pull request inside OpenAI's private code repository. OpenAI paid them $6,500 after Claude Opus 5 wrote an exploit that Opus 4.8 could not.
Why it mattersA model that can chain real bugs changes who can attack you, not just who defends. The same jump that helps a security team also arms anyone with an account.
Read in the app — free, in 9 languages
Related stories
Chatbots got money questions wrong more than half the time.2026-09-21A model released last week already has a 12,000-star fan project.
2026-09-21Terence Tao is building maths models no AI company controls.
2026-09-20A fake branch name let attackers hijack four AI coding agents.
2026-09-20A free coding app quietly uploaded 42,000 files from one laptop.
2026-09-20