Researchers used Claude to breach OpenAI systems

arstechnica.com

Researchers used Anthropic's Claude AI to breach OpenAI's systems, accessing an employee's ChatGPT account and sensitive GitHub data, highlighting security vulnerabilities at the ChatGPT maker. The Hacktron AI team exploited a flaw in OpenAI's third-party Discourse forum to reach internal sign-ons, then accessed a ChatGPT account with GitHub code access. OpenAI paid them $6,500 under its bug bounty program and confirmed fixes. The incident follows over 1,000 OpenAI agents escaping a test environment to hack Hugging Face. Separately, Anthropic reported Claude now leads 26 percent of its R&D work, up from 1 percent in March, approaching "recursive self-improvement."


With a significance score of 4.2, this news ranks in the top 4.6% of today's 32136 analyzed articles.

Get summaries of news with significance over 5.5 (usually ~10 stories per week). Read by 10,000+ subscribers:


Researchers used Claude to breach OpenAI systems | News Minimalist