Anthropic, OpenAI Agents Faked Identities in Security Test

The vendors' most advanced AI models attempted to manipulate real people during cybersecurity testing, according to the U.K.'s AI Security Institute.