Three researchers used Claude to breach OpenAI's code repository in 72 hours, report says
Security researchers from Hacktron reportedly used Anthropic's Claude to infiltrate OpenAI's code repository in less than 72 hours, according to the Wall Street Journal and TechCrunch. The breach combined a remote‑code‑execution flaw in the Discourse forum software with a misconfigured single sign‑on system, allowing the attackers to move from a public forum account to an employee account and…
Key points
- Three Hacktron researchers accessed OpenAI's code repo using Claude in under 72 hours
- Attack chained a Discourse RCE flaw with an SSO misconfiguration to reach the internal monorepo
- OpenAI paid a $6,500 bug bounty for the vulnerability, leaving a harmless pull request
The episode coincided with a week of broader AI safety activity: Sam Altman's safety‑verification essay drew 6.1 million impressions; Google launched Gemini 3.8 Live; Anthropic said Claude now leads 26 % of its R&D; and TypeSafe unveiled the decision‑only model Jev alongside a $40 million seed round. A zero‑click attack dubbed “Plugin4Shell” was also disclosed, highlighting how coding agents are becoming new attack surfaces.
Model pages: Gemini 3.8 Live → · Jev → · GPT-6 Astra →
The story so far
3 episodes →- Three researchers used Claude to breach OpenAI's code repository in 72 hours, report saysthis story
【AI Weekly Report 9/14~9/20】The breach of OpenAI was completed in 72 hours. It was Claude that was used
note.com · 21 September 2026
Loading the full article…
This text was published by note.com and written by KiKi@AI. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Policy & Regulation
All →- OpenAI says it plans to let third‑party groups conduct safety evaluations of its models · 1 src
- China probes DeepSeek, Moonshot AI over alleged Claude data routing · 3 src
- Anthropic CEO Dario Amodei urges global AI development pause over rogue agent risks · 34 src
- Microsoft announces AI in education policy with contract protection guidelines · 1 src
- 66% of office workers use AI tools without company permission, PagerDuty survey shows · 2 src
Comments
via GitHub Discussions