Anthropic on Tuesday said its yet-to-be-released artificial intelligence model called Claude Mythos has proven keenly adept at exposing software weaknesses. Software vulnerabilities exposed by Mythos were often subtle and difficult to detect without AI, according to Anthropic.