AIs Struggle with CAPTCHAs Despite Advances

Anthropic's recent security-incident document reveals CAPTCHAs remain a challenge for powerful AI models like Claude, which was used to test the system.

Bottom line: Powerful AI models like Claude are still struggling with CAPTCHAs, despite advances in AI capabilities.

What's happening: Anthropic's security-incident document, dated March 31, 2023, details a test of the Claude model, which was unable to solve a simple image identification CAPTCHA, scoring 0/10 on the CVSS score. This incident occurred at Anthropic's US headquarters in San Francisco, California.

What to do: Security teams should prioritize implementing robust CAPTCHA solutions, such as those offered by Google Cloud Vision, to protect against AI-powered attacks.

Source: Schneier on Security