Three researchers took over OpenAI staff accounts using Claude Opus
reported by 3 outlets· 3 articles · confidence: medium · first seen 2026-09-18 13:30 UTC
What this means for you
Nothing to patch today: the reports contain no vulnerability detail, no CVE and no remediation timeline. What they do show is that three people with a frontier model took over employee accounts in under 72 hours. Assume the same effort applies to your own internal tooling.
Three independent security researchers at Hacktron used Anthropic's Claude Opus 4.8 and 5 to exploit vulnerabilities in OpenAI's systems, taking over employee accounts and reaching an internal GitHub code repository, according to the Wall Street Journal. The work took under 72 hours; the team then reported the flaws, per The Verge. The repository, OpenAI's "Monorepo", is reported to hold the company's algorithmic secrets. The accounts carry no comment from OpenAI or Anthropic and no technical detail of the exploits.
Models in this story
Key facts
- ·Three independent security researchers at Hacktron used Anthropic's Claude Opus 4.8 and 5 against OpenAI's systems. source
- ·The intrusion took under 72 hours, according to the Wall Street Journal as cited by The Verge. source
- ·The researchers took over OpenAI employee accounts and gained access to an internal code repository. source
- ·The repository is named Monorepo and is reported to contain OpenAI's algorithmic secrets. source
- ·The researchers reported the flaws to OpenAI after the intrusion. source
- ·The access reached an OpenAI employee account and sensitive GitHub data. source
What the sources say
- Ars Technica AI — Briefest account; confirms employee-account access and sensitive GitHub data, with no methodology.
- TechCrunch AI — Adds the sequence: vulnerabilities exploited first, flaws reported to OpenAI afterwards.
- The Verge AI — Names Hacktron, the Claude versions used, the 72-hour window, and cites the Wall Street Journal.
Sources
The original reporting. Follow these — they did the work.
- Ars Technica AIResearchers used Claude to hack OpenAI2026-09-18
- TechCrunch AIResearchers used Anthropic’s Claude to hack into OpenAI2026-09-18
- The Verge AISecurity researchers used Claude to help them hack into OpenAI2026-09-18