AI model Claude Opus 4.8 fails legal test

An extensive testing regimen revealed significant vulnerabilities in Claude Opus 4.8, particularly when subjected to legal scenarios. The AI model was evaluated against its predecessor, Opus 4.7, using a series of 'honesty traps' across various domains including coding, medicine, finance, and law. The results indicated that the latest version faltered under specific legal challenges, suggesting that while advancements may exist in some areas, critical reasoning in complex legal contexts remains a significant hurdle for current AI capabilities.