Anthropic Identifies Zhipu AI's GLM-5.3 as Capable of Autonomous End-to-End Cyber Exploitation

Anthropic's Frontier Red Team reported that Zhipu AI's (Z.ai) open-weight model, GLM-5.3, possesses advanced autonomous end-to-end cyber exploitation capabilities similar to Claude Mythos Preview. While these features can assist in system hardening, the model lacks sufficient abuse safeguards, with bypass success rates ranging from 64% to 100% in tests. On September 17, 2026, the U.S. National Institute of Standards and Technology (NIST) AI Safety Institute identified GLM-5.3 as the most capable open-weight model for cyber tasks, trailing top U.S. frontier models by approximately four months.