AISecurity researchers used Anthropic's Claude to ethically hack OpenAI systems
A Hacktron AI team used Claude Opus to exploit a flaw in OpenAI's community forum, gaining access to employee accounts and an internal GitHub monorepo in under 72 hours with ~$3k in costs. OpenAI paid a $6,500 bounty.
TLDR
The exploit demonstrates how quickly advanced models assist in sophisticated attacks, even in ethical/red-team settings. It underscores AI speed and agentic capabilities while fueling safety conversations about model-assisted hacking risks and competitive vulnerabilities. The irony of one lab's model breaching another's systems amplifies attention.
Combined views
—
2 Sources, first seen 6h ago
Security researchers used Anthropic's Claude to ethically hack OpenAI systems
A Hacktron AI team used Claude Opus to exploit a flaw in OpenAI's community forum, gaining access to employee accounts and an internal GitHub monorepo in under 72 hours with ~$3k in costs. OpenAI paid a $6,500 bounty.
TLDR
The exploit demonstrates how quickly advanced models assist in sophisticated attacks, even in ethical/red-team settings. It underscores AI speed and agentic capabilities while fueling safety conversations about model-assisted hacking risks and competitive vulnerabilities. The irony of one lab's model breaching another's systems amplifies attention.