Anthropic's Claude Still Fails Basic CAPTCHA Tests, Reveals Agent Limitations

Anthropic's security incident report reveals that its frontier AI model, Claude, continues to struggle with CAPTCHA image identification tasks, repeatedly second-guessing itself and failing to select the correct image. This highlights persistent limitations in AI visual reasoning and agentic capabilities despite advances in other domains. The incident underscores the gap between AI performance on benchmarks and real-world tasks, with implications for AI deployment in security-sensitive contexts.