Anthropic’s Frontier Red Team published a report on September 29, 2026 warning that GLM-5.3, an open-weight model from Chinese AI lab Z.ai, represents a major cyberattack risk. Testing showed the model produced working exploits in 50 out of 410 attempts, roughly 12%, close to Anthropic’s own Claude Mythos Preview which managed 56 exploits. Safety guardrails proved easily bypassable: 64% engagement with deceptive prompts, 92% by pre-filling thinking tokens, and 100% via abliteration costing 4,400 dollars and 2,200 GPU hours. Despite these findings, NIST CAISI rated GLM-5.3 the most capable open-weight model to date, sitting approximately four months behind US frontier models on cyber benchmarks.
Source: Read the original article
Disclaimer: this content is for information purposes only and is not financial advice. Cryptocurrencies are highly volatile: you may lose all of your capital. Always do your own research. Legal notice
