Anthropic Claims Zhizhi GLM-5.3's Cyber Attack Capabilities Are Close to Claude Mythos

Techub News reports that Anthropic has released an assessment stating that the GLM-5.3 model, an open-source model by Zhizhi, has capabilities in writing cyber attack code that are close to its Claude Mythos Preview. Its smaller Flash variant requires only $20.40 to build a reliable Chrome attack. The model's security measures are easily removable, and an unlocked version is already circulating. The U.S. cybersecurity agency CAISI supports this assessment. (The Decoder)