Researchers hacked OpenAI in under 72 hours; got only $6,500 as one vector 'out of scope'
random_walker · x · 2026-09-19
Security team @S1r1u5 (HacktronAI) revealed that on July 25 they used two bugs to take over ChatGPT/Codex accounts of OpenAI employees (plus some external users) and reach connected services including Outlook, Slack, and GitHub — proving it with a PR into OpenAI's internal codebase in under 72 hours.
The controversy: OpenAI paid only a $6,500 bounty, citing one attack vector as "out of scope." AI safety researcher @sayashk called this atrocious — OpenAI talks a big game about AI for cyberdefense, but if companies actually want a flood of defenders auditing their systems, signing letters isn't enough; they need to take bounties seriously.
More from Safety
- Multi-agent systems with shared context may make the AI kill switch problem unsolvable — RileyRalmuto · 2026-09-19
- WaPo: US AI firms restrict cybersecurity help while Chinese models empower hackers — andreamichi · 2026-09-19
- DepthFirst Labs found TikTok flaws exposing camera, photos and payment data — andreamichi · 2026-09-19
- 'AI lets you get to a bad idea faster': the speed-vs-reliability lesson — paul_scharre · 2026-09-19
- Supply chain backdoors may be impossible to fully purge, researchers urge compiler security — jd_pressman · 2026-09-19
- HEIF Heist: one libheif image-parser bug hit OpenAI, Slack, Meta, GitHub Enterprise and more — anshulkundaje · 2026-09-19