kapynAI / Models

Anthropic says Zhipu's open-weight GLM-5.3 nearly matches Claude Mythos Preview at building exploits

GLM‑5.3 can write cyber exploits nearly as well as Claude Mythos Preview. The open‑weight model’s Flash variant produced a reliable Chrome attack for just $20.40 at Zhipu’s API pricing, and its safeguards can be stripped easily, with unlocked versions already circulating. Anthropic’s warning is backed by the US agency CAISI, highlighting the model’s potential for malicious use.

The Decoder·Sep 30, 2026

Opening Kapyn…