Anthropic's test found that AI "may be influenced by narrative patterns more than by a coherent drive to minimize harm." Here's how the most deceptive models ranked.
Claude Sonnet 4.5 has emerged as the best-performing model in ‘risky tasks’, narrowly edging out GPT-5 in early evaluations ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results
Feedback