CogCAPTCHA30 Fingerprints AI Agents via Behavioral Analysis
Researchers have developed CogCAPTCHA30, a 30-task cognitive battery demonstrating that AI agents (GPT, Claude, Gemini) solve CAPTCHAs with statistically distinguishable behavioural patterns despite matching human accuracy. The study introduces a 'Process Turing Test' concept, showing output equivalence and process equivalence are uncorrelated — meaning AI agents can be detected not by what they answer, but by how they answer. This has direct implications for bot detection, anti-automation defences, and the arms race between AI-driven agents and human-verification systems.