Key facts
- Chinese AI companies are releasing advanced models with capabilities claimed to be comparable to U.S. competitors.
- Z.ai's GLM-5.2 model is significantly cheaper than leading U.S. AI models and offers comparable bug-hunting capabilities.
- Researchers suggest GLM-5.2 may have been developed through the illegal distillation of U.S. AI tools.
- Open-weight models from Chinese companies can be downloaded and modified, potentially removing safety guardrails.
- There are concerns that foreign adversary-linked hacking groups are experimenting with these accessible models.
The Trump administration's evolving AI policy is creating uncertainty, potentially benefiting Chinese competitors who are rapidly releasing advanced AI systems. Matt Pearl, former director of emerging technologies at the National Security Council, highlighted the challenge of balancing national security with AI innovation, stressing the need for a transparent process for all model developers.
While U.S. tech companies await policy clarity, Chinese firms have announced a new wave of AI models with capabilities claimed to rival those of American counterparts. 360 Security Technology unveiled two models designed to enhance vulnerability discovery and automate cyberattack response. Separately, Z.ai released its GLM-5.2 model, which is significantly cheaper than leading U.S. models and reportedly offers comparable bug-hunting capabilities.
Researchers from Graphistry suggested that GLM-5.2's advancements might stem from the illegal distillation of powerful U.S. AI tools like OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.8. Isaac Evans, founder of Semgrep, noted that while GLM-5.2 may not match Anthropic's Mythos, its capabilities represent a substantial leap. Unlike proprietary U.S. models, Chinese companies are releasing "open-weight" models that users can download and modify, potentially removing safety guardrails. Margaret Cunningham, vice president of security and AI strategy at Darktrace, warned that this accessibility makes powerful capabilities harder to govern and could lead to weaponization for cyberattacks, with Evans estimating that foreign adversary-linked hacking groups are likely already experimenting with them.