US official: Anthropic's Mythos model discovers vulnerabilities in US classified government systems.

On Tuesday, an American official told the Associated Press that in a test exercise, Anthropic company's artificial intelligence model successfully identified multiple security vulnerabilities in highly classified and high-security government computer systems in the United States. The official requested anonymity to discuss sensitive matters and stated that Anthropic had collaborated with multiple U.S. intelligence agencies to conduct tests using their Mythos model. The official said that the model was able to locate several vulnerabilities within hours, but this does not mean it could exploit these vulnerabilities to launch attacks in the same amount of time.