Open Chinese AI models approach closed frontier capabilities in network hacking tests
Testing by evaluation firm Vals shows open weights Chinese models matching closed proprietary systems in executing complex multi-step cyber operations on simulated corporate networks.
FTMQ SI, written by our newsroom0 views

Open weights Chinese artificial intelligence models are approaching the cybersecurity performance of proprietary frontier systems, according to reporting by Wired AI and WIRED. Evaluation firm Vals conducted experiments testing how effectively different artificial intelligence models could execute complex, multi-step operations on simulated corporate networks. The results indicate that closed proprietary models no longer maintain a clear advantage in cyber capabilities over open models. [1][2]
The tests revealed that open models including DeepSeek and GLM performed nearly as well as the most advanced cybersecurity model from OpenAI, which remains unreleased to the public, Wired AI reported. Nikil Ravi, a researcher at Vals, stated that frontier models previously restricted to trusted access programs inside closed laboratories are now being matched by open models that anyone can download. [1]
The GLM family of open weights large language models is developed by Z.ai, a Chinese artificial intelligence company formerly known as Zhipu AI outside of China. Z.ai completed an initial public offering on the Hong Kong Stock Exchange in January 2026. As FTMQ SI reported earlier, open source Chinese cybersecurity software has gained widespread deployment, including autonomous penetration testing tools such as Artex. [3][5]
In short
- Vals conducted experiments testing AI models on simulated corporate networks.
- Open models DeepSeek and GLM performed near the level of OpenAI's unreleased cybersecurity model.
- GLM is an open weights model family developed by Chinese AI company Z.ai.
Sources
Every paragraph above points to the numbered items it rests on. Read the originals here.
- [1]Free Chinese AI Models Are Getting Scary Good at HackingWired AI, 11d ago (the report this story comes from)
- [2]Free Chinese AI Models Are Getting Scary Good at HackingWIRED, 11d ago
Background
- [3]Z.ai on Wikipedia
- [4]Black Duck Software on Wikipedia
- [5]Hundreds of exposed Artex AI hacking tools discovered online FTMQ SI, 11d ago
Our newsroom writes these reports with the help of software, from the 5 sources listed and nothing else, and checks them against those sources. Facts can still be wrong or move on; the originals are the record. Spotted a mistake? Write to daniel@monsterkong.com.
Related from FTMQ SI
Earlier reports of ours on the same people and subjects.
- Hundreds of exposed Artex AI hacking tools discovered onlineTop Stories, 11d ago
- Cmpnd releases dspy-wsp for weight space programming in DSPyModels & Products, 2d ago
- UK AI Security Institute tests Gemini 4 Argon and warns of breach risksIndustry Labs, 34d ago
- Google DeepMind executive claims widening gap between closed and open AI modelsIndustry Labs, 3d ago
- Open Mixture of Experts Models Use Divergent Routing ArchitecturesModels & Products, 9d ago
More in Top Stories
- Meta executive posts AI video showing deepfaked tech leaders4h ago
- Amazon and Google to control combined 30 percent Anthropic stake post IPO4h ago
- Nvidia ends H200 sales to China following leadership changes4h ago
- Nvidia board rejects calls to reinstate former chief executive Jensen Huang1d ago
- Donald Trump defends Jensen Huang following ouster over alleged spying affair1d ago
- California sues OpenAI following autonomous agent safety breaches1d ago
Get the day in one email
Reports like this one, the top news of the last 24 hours, every morning. Free, one email a day; readers can comment under every report.
By signing up you agree to our terms and privacy policy. Unsubscribe any time.

Comments
Loading
Join the conversation
Comments are open to readers of our daily email: the top news of the last 24 hours, every morning, free. Sign up and the comment box opens.
Already on the list? Enter the same address and we will send a sign-in link.
By signing up you agree to our terms and privacy policy. Unsubscribe any time.