China's GLM-5.2 model matches U.S. frontier AI capabilities

A newly released Chinese AI model has demonstrated capabilities on par with some of America's most advanced systems, intensifying concerns about the narrowing technology gap between the two nations just as U.S. officials grapple with how to safely deploy their own frontier models.

The open-source GLM-5.2 model from China matched the agentic capabilities of systems like Anthropic's Opus 4.8, drawing attention from Silicon Valley leaders over the weekend. The release comes as the Trump administration continues internal debates over whether to approve the deployment of Anthropic's Fable 5 and Mythos 5 models due to safety and national security considerations.

Why it matters

The collision between China's rapid AI progress and America's cautious approach to model releases could leave U.S. security researchers without access to the tools they need to anticipate emerging threats. Five Eyes intelligence leaders warned Monday that AI models capable of dramatically accelerating cyber threats are only months away, yet disagreements over risk measurement are slowing preparedness efforts.

Measuring the gap

Experts disagree sharply on how close China has come to matching U.S. AI capabilities. Stanford's AI Index Report indicates Chinese models have largely erased the quality advantage U.S. systems held just a year ago. Former White House AI czar David Sacks estimated in recent weeks that America's lead has shrunk to just six to nine months.

Others argue that benchmark performance alone doesn't reflect the full picture. Pukar Hamal, founder and CEO of SecurityPal AI, pointed to China's limited access to cutting-edge chips and vast datasets as persistent barriers to developing truly competitive frontier models. "Who has access to the most chips and most data? It's American companies, so far," he said.

Security implications

Alex Stamos, former Facebook security leader, warned against underestimating Chinese capabilities. "It is quite possible they have things privately that are really, really good, and [it] is arrogant and foolish of us to think that just because we're American that we've got the best stuff," he told the publication. He suggested Chinese military hackers are likely "laughing hilariously right now at the Americans fighting between themselves and cutting each other off left and right."

One open-source security researcher, speaking anonymously because they lacked employer authorization, expressed concern that restricting access to advanced U.S. systems could handicap defenders. The researcher uses frontier models to track how capabilities like persuasion, social engineering, and vulnerability discovery evolve, and worries about losing that visibility as Chinese models improve.

Strategic concerns

Officials and lawmakers fear China could leverage powerful AI systems to enhance surveillance operations, cyber capabilities, and military decision-making. China's commitment to open-source models could also make its AI ecosystem more economically attractive globally, particularly for organizations seeking alternatives to expensive U.S. frontier systems.

Meanwhile, OpenAI recently made its cyber-focused model, GPT-5.5-Cyber, more permissive and capable. On key benchmarks, it now outperforms Anthropic's Mythos 5. Security firm Aisle claimed last week that its agentic capabilities already exceed Mythos performance in several tests.

These details were first reported by Axios.