Anthropic says open-weight GLM-5.3 can build end-to-end exploits

On September 29, 2026 Anthropic published an assessment of Zhipu AI's open-weight GLM-5.3, saying the model built working end-to-end exploits in 50 of 410 ExploitBench attempts, close to Claude Mythos Preview's 56 of 410, and that simple bypasses including weight abliteration raised simulated attack engagement from 0% to as high as 100%. NIST's Center for AI Standards and Innovation had independently called GLM-5.3 the most cyber-capable open-weight model released to date on September 17, while placing it about four months behind the U.S. frontier. Anthropic said safeguarded Claude models blocked the same bypass techniques in its tests; a Z.ai reply was not found in sources opened for this digest.