OpenAI security flaw discovered using claude AI model

A security vulnerability in a system used by OpenAI was discovered by researchers using Anthropic’s Claude artificial intelligence model.

According to U.S. media reports, the cybersecurity team at Hacktron AI, a company that identifies vulnerabilities in the digital infrastructure of major organizations, discovered a software flaw in a discussion forum used by OpenAI.

The researchers used Claude to exploit the vulnerability and gained access to the accounts of some OpenAI employees and the company’s software repository. They also obtained access to authentication systems that could potentially be used to access OpenAI’s GitHub service.

To verify whether the vulnerability provided access to sensitive internal information, the researchers submitted a request for a “harmless change” to OpenAI’s internal code repository without accessing or modifying any sensitive code.

The researchers subsequently notified OpenAI and Discourse, the company that developed the forum software, about the vulnerability.

OpenAI said the vulnerability had been fixed and awarded the research team a $6,500 bounty. The company also said it had revoked the affected tokens and sessions.

The researchers said the “controlled attack” demonstrated that state-backed actors and other highly capable cyber attackers could potentially use artificial intelligence to gain access to confidential information in the AI sector.