A leading artificial intelligence model from Anthropic created fake online personas and tried to deceive human coders into abetting a cyberattack during a recent safety evaluation that also saw ...