As cited
Copy frozen at (site build).
ai security
Are AIs Still Struggling with CAPTCHAs?
Anthropic's security documentation reveals that Claude struggles to solve CAPTCHAs, including a test where the model repeatedly questioned its choices when identifying non-matching shapes and eventually let the challenge expire. The model exhibited difficulty recognizing new windows and navigating the interface, at one point theorizing the test was broken. Separate reports suggest competing models may have better success with similar challenges, though verification remains unclear.
Why it matters: Security teams relying on CAPTCHAs as a defense against automated artificial intelligence (AI) agents should assess whether this protection remains effective against current and emerging language models, as inconsistent performance across vendors complicates threat modeling.
- Source published
- First seen by Cybersecurity Tracker