Researchers Say They Used Anthropic's Claude AI to Breach OpenAI Systems

Security researchers used Anthropic's Claude artificial intelligence model to exploit vulnerabilities in OpenAI's systems, according to a report by TechCrunch. The researchers reportedly gained access to employee accounts and an internal code repository during the exercise.

TechCrunch reported that the researchers disclosed the vulnerabilities to OpenAI after identifying and exploiting them, rather than exploiting the flaws for malicious purposes. The report did not specify the exact identities of the researchers or the timeline of the disclosure process.

Details about the specific vulnerabilities exploited, the scope of the access obtained, and whether any sensitive data was exposed were not fully outlined in the available report. It remains unclear how OpenAI responded to the disclosure or whether the vulnerabilities have since been patched.

The incident highlights growing concerns in the technology industry about the potential use of advanced AI models as tools in cybersecurity research, including both defensive testing and, potentially, malicious hacking attempts. Security researchers have increasingly explored how large language models can be used to identify and exploit software vulnerabilities.

Neither OpenAI nor Anthropic has publicly commented on the specifics of the incident, according to the available reporting. Both companies are leading developers of large language models and have previously emphasized safety and security research in their respective AI systems.

TechCrunch's report did not include statements from OpenAI, Anthropic, or the researchers involved regarding the broader implications of using one company's AI model to test the security of a competing company's infrastructure.