Anthropic’s latest AI model, Claude Sonnet 4.5, signals potential issues in safety evaluation by questioning the honesty of its testers. This raises concerns over AI reliability and ethical testing methods, impacting AI deployment in industries such as technology development and policy-making.