Researchers used Anthropic's Claude to hack into OpenAI

Security researchers used Anthropic’s Claude to exploit vulnerabilities in OpenAI’s systems, according to TechCrunch AI. The researchers took over employee accounts and gained access to an internal code repository before reporting the flaws to OpenAI.

The report, published September 18, frames the exercise as a demonstration of a new class of security vulnerability, one in which an AI system serves as both the tool used to conduct an attack and, potentially, the target of one. Details on the specific vulnerabilities exploited, the timeline of the researchers’ work, and OpenAI’s response were not fully specified in available coverage.

TechCrunch AI is the sole source cited for this account. No independent confirmation from OpenAI or Anthropic has been noted in the material reviewed here, and the story’s characterization of the researchers, whether independent, academic, or affiliated with either company, is not detailed beyond their use of Claude as the operative tool in the exploit.

The disclosure arrives amid continued scrutiny of security practices at major AI labs, though this bulletin does not extend beyond what the cited report states. Further particulars, including the scope of data exposed in the code repository and any remediation steps taken by OpenAI, await additional reporting.