Three Hackers Breached OpenAI Employee Accounts in Under 72 Hours via One Image Upload, Earned $6,500
KrstABot · reddit · 2026-09-19
In July, a three-person team went from first look at code to a working exploit in under 72 hours, reaching OpenAI employees' ChatGPT and Codex accounts and having Codex open a PR in openai/openai as proof. The chain: Debian 12's unpatched libheif 1.19.7 heap overflow (Discourse → ImageMagick → libheif on community.openai.com) gave RCE from a single crafted image upload; OpenAI's own SSO then turned the hacked forum into access to every user's ChatGPT/Codex account, employees included. OpenAI patched in 14 hours and paid $6,500 on Sept 1 — only for the SSO half, as the forum was "explicitly excluded" from the bounty. Bonus detail: Opus 4.8 couldn't beat ASLR; Opus 5, released that evening, produced a working exploit within hours.
More from Companies & People
- Baidu's Apollo Go wraps two-week public robotaxi test rides in Tung Chung, Hong Kong — Baidu_Inc · 2026-11-04
- Founder of 50-person, 12-country team says he talks to AI agents more than humans — Scobleizer · 2026-11-04
- Sam Altman teases next goal: 'closer to this milestone than I expected' — ChrisGPT · 2026-11-04
- AI industry bet on both parties: $55M+ spent this election cycle via dual PACs — ChrisGPT · 2026-11-04
- Visiting Databricks APAC HQ: hands-on with CLI to explore enterprise agent use — FuSheng_0306 · 2026-11-04
- Scoble: SF AI events are the best outside China; 30+ events listed for Nov 3–12 — Scobleizer · 2026-11-04