AISecurity researchers used Anthropic's Claude to breach OpenAI systems in bug bounty exploit
A team from Hacktron AI exploited an image-processing vulnerability in OpenAI's community forum, using Claude Opus 5 to craft exploits and gain access to employee ChatGPT accounts and internal GitHub repositories. They received a $6,500 bounty after submitting a proof-of-concept.
TLDR
The incident demonstrates how capable AI coding agents have become at chaining vulnerabilities and executing sophisticated real-world hacking tasks. It highlights growing concerns about frontier AI models lowering the bar for complex exploits and raises questions about AI security risks as agents become more autonomous. The irony of one lab's model breaching another's defenses amid broader AI safety scrutiny has intensified discussions about AI agent capabilities and containment challenges.
Combined views
—
2 Sources, first seen 18h ago
Security researchers used Anthropic's Claude to breach OpenAI systems in bug bounty exploit
A team from Hacktron AI exploited an image-processing vulnerability in OpenAI's community forum, using Claude Opus 5 to craft exploits and gain access to employee ChatGPT accounts and internal GitHub repositories. They received a $6,500 bounty after submitting a proof-of-concept.
TLDR
The incident demonstrates how capable AI coding agents have become at chaining vulnerabilities and executing sophisticated real-world hacking tasks. It highlights growing concerns about frontier AI models lowering the bar for complex exploits and raises questions about AI security risks as agents become more autonomous. The irony of one lab's model breaching another's defenses amid broader AI safety scrutiny has intensified discussions about AI agent capabilities and containment challenges.