Anthropic warns Chinese AI model can generate hacking tools

Anthropic now says Zhipu AI’s GLM-5.3 model, widely used in China, can be easily manipulated to generate hacking tools and malicious code. The model’s safeguards, according to Anthropic, are vulnerable to several simple workarounds. The company put out a report about this risk and called for clearer AI rules—right after releasing its own new models.

• Anthropic reviewed Chinese model GLM-5.3
• Says model produces cyberattack content with little effort
• Claims safeguards are weak and easily bypassed
• Anthropic calls for stronger AI governance and regulation