Free Chinese AI Models Approach Top Proprietary Systems in Simulated Hacking Tests

An experiment tested AI models on their ability to break into simulated computer networks. Chinese models DeepSeek and GLM performed closely to OpenAI's most advanced cyber-capable model, which has not been released publicly. The results suggest that openly available AI systems are narrowing the gap with leading proprietary models in offensive cybersecurity tasks. The findings raise concerns among security observers about how such capabilities could be misused. The article is based on a single experiment involving simulated environments rather than real-world attacks.