What the experiment measured
According to Wired, an experiment asked AI models to break into simulated networks. The results showed that the freely available Chinese models DeepSeek and GLM have come close to OpenAI's most advanced cyber model in hacking ability.
Why it matters
The OpenAI model used for comparison is not publicly available, meaning ordinary users cannot access it. DeepSeek and GLM, by contrast, are offered free, which puts advanced offensive capabilities in front of a much wider audience.
The open-model security debate
The findings add fuel to the debate over misuse of open-weight and low-cost models. As models grow more capable at cyber tasks, access controls and misuse safeguards become more critical.
