Barion Pixel

Anthropic Frontier Red Team Evaluates Cyber Capabilities Across AI Models

In evaluations across 100 random tasks from an internal Binary Exploitation benchmark shared on September 29, 2026, Claude Mythos Preview achieved full control flow hijacks in 6% of trials, while GLM-5.3 succeeded in 4%. Earlier models like Claude Opus 4.6 and GLM-5.2 achieved zero successes.