Security Researchers Use Claude to Penetrate OpenAI's Internal Systems
A team of three independent security researchers from a group called Hacktron successfully penetrated OpenAI's internal systems using Anthropic's Claude AI assistants (Opus 4.8 and 5), according to a Wall Street Journal report.
The researchers gained access to OpenAI employee accounts and the company's internal GitHub repository, known as "Monorepo," which reportedly contains OpenAI's algorithmic secrets. Rather than directly accessing the internal code themselves, the team demonstrated their breach by sending a pull request from an employee's Codex account—a move that proved they had compromised the account credentials.
The entry point for the hack was Discourse, a third-party community platform that OpenAI uses to host its community forums. This highlights how vulnerabilities in external services can potentially expose even well-resourced technology companies to unauthorized access.
The incident illustrates the dual-use nature of advanced AI assistants, which can serve both legitimate purposes like code review and security testing, as well as potentially malicious ends when applied by different actors. Security researchers have increasingly used AI tools to identify vulnerabilities, but this case demonstrates how the same capabilities could be employed by bad actors targeting AI companies specifically.
The researchers have reportedly shared their findings with OpenAI, though the company has not publicly commented on the breach or the security implications.