Are AIs Still Struggling with CAPTCHAs?

A recent security incident document from Anthropic mentions that CAPTCHAs are still a challenge for the Claude model, a powerful AI. The document includes a transcript where Claude struggles to solve a simple image identification test, unable to decide which image to select. Claude repeatedly reviewed the same images and questioned its own conclusions, indicating difficulty with the task. The test involved identifying a shape that didn't match the others displayed, but Claude was unable to make a decision. The transcript shows Claude's thought process, with it expressing uncertainty and revisiting the same images. This suggests that CAPTCHAs can still be an obstacle for advanced AI models like Claude.