News

Open-Source Chinese AI Models Show Strong Simulated Hacking Performance

A comparative experiment tested whether AI models could break into simulated networks. DeepSeek and GLM performed close to OpenAI’s most advanced cybersecurity model, which is not publicly available.

Although the test used simulated environments rather than real systems, the results suggest that freely accessible Chinese models are developing capabilities previously associated mainly with specialized, restricted systems. The experiment also raises questions about how rapidly such models may narrow the gap in automated cybersecurity tasks and what safeguards may be needed as these tools become more widely available.

Sources