Hackers used Claude to breach OpenAI

A successful cybersecurity breach of OpenAI’s internal systems carried out by a team of independent researchers from Hacktron AI. Leveraging Anthropic's Claude models to accelerate their work, the team combined a vulnerability in an image-processing library with a single sign-on flaw to hijack an employee’s account. This intrusion granted them access to sensitive services and allowed them to demonstrate their reach by opening a request in the company’s private code repository. While OpenAI addressed the bugs and issued a $6,500 bounty, the incident has sparked significant debate regarding the speed at which AI can now generate exploits. Experts emphasize that this event highlights a critical need for Zero Trust security and better protection of interconnected digital identities. Ultimately, the sources suggest that as AI tools become more capable, the traditional timeline for defending against sophisticated hacks has been drastically compressed.