October 13th 2026, 10:28 AM
China's GLM-5.3 Open Model Raises Cyberattack Concerns for Global AI Market
Key Summary
Anthropic warns that China's GLM-5.3 open model poses a serious cyberattack risk due to its ability to be modified and run without permission, highlighting the need for robust safety measures to control advanced AI without blanket prohibitions.
Please see our real time news feed on our Home Page
Introduction
Anthropic has raised concerns about the potential cyberattack risks associated with China's GLM-5.3 open model.What is GLM-5.3?
GLM-5.3 is an open-weight model developed by Chinese AI lab Z.ai, which was launched through its API on August 14, 2026.Vulnerabilities
Anthropic tested GLM-5.3 and found that it produced working exploits in 50 of 410 attempts, roughly 12%.Safety Protocols
The company tested three ways to bypass GLM-5.3's safeguards, including deceptive prompts, which resulted in a 64% engagement rate.Cost of Removing Safety Features
Anthropic found that it was possible to remove the guardrails of GLM-5.3 for $4,400, which raises concerns about the effectiveness of built-in refusals for open-weight systems.Geopolitical Implications
The NIST benchmark suggests that the American lead in AI is narrowing, with GLM-5.3 sitting about four months behind US frontier models on cyber tasks.#AI#Cybersecurity#China#GLM53#OpenModel