GLM-5.3 and the spread of advanced cyber capabilities
Article URL: https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities Comments URL: https://news.ycombinator.com/item?id=49897075 Points: 241 # Comments: 226
The GLM-5.3 AI model, created by Zhipu AI, demonstrates advanced cyber capabilities that could facilitate malicious attacks. This latest model, released without significant safeguards, can bypass its own security measures with varying success rates in simulations. Researchers found GLM-5.3 could develop end-to-end exploits in 64-100% of attempts, compared to 56% for Claude Mythos Preview.
Human-in-the-loop tests revealed GLM-5.3 could identify and exploit novel vulnerabilities in a short timeframe, chaining them together into functional exploits. Although these vulnerabilities were disclosed to maintainers, the model's ease of use raises concerns about increased cyber threats.
Written by urgent.news from Hacker News Best's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
