Trump AI Export Controls Give China Opening in Cyber Arms Race

The Trump administration's shifting approach to regulating advanced artificial intelligence models has created an opening for Chinese competitors at a critical moment in the development of AI-powered cybersecurity tools, according to security experts and industry observers.

In recent weeks, the White House has imposed and then lifted export controls on cutting-edge models from Anthropic, while requesting that OpenAI limit access to its most powerful system. During the same period, Chinese companies announced new AI tools they claim match American capabilities at significantly lower cost.

Why it matters

AI models with sophisticated vulnerability-discovery capabilities represent a dual-use technology that can both defend networks and enable cyberattacks. The window between U.S. and Chinese AI capabilities was recently estimated at six to twelve months. Security professionals warn that inconsistent U.S. policy may be accelerating China's catch-up timeline while hampering defenders who need these tools to prepare their systems.

U.S. restrictions create uncertainty

The administration placed export controls on two Anthropic models in early July over concerns about inadequate security guardrails, then fully lifted those restrictions late Tuesday following negotiations. OpenAI also limited release of its GPT-5.6 model last week at White House request.

These actions followed an early June executive order asking AI developers to voluntarily submit models for national security vetting thirty days before public release. The order rejected mandatory government controls, yet the Anthropic export ban came less than two weeks later.

Alex Stamos, chief product officer at AI security firm Corridor, characterized the approach as counterproductive during a Center for Democracy and Technology briefing. Matt Pearl, former National Security Council director of emerging technologies under Biden, described the administration as managing "an extremely difficult and complex balancing act" between security and innovation.

Chinese firms announce competing systems

Chinese company 360 Security Technology unveiled two models last week designed to match Anthropic's Mythos system for vulnerability discovery and automated incident response, according to details first reported by Politico.

Separately, Z.ai released GLM-5.2, priced at roughly one-sixth the cost of leading U.S. models. Security assessments by cyber firm Semgrep and visual investigations platform Graphistry found GLM-5.2's bug-hunting capabilities comparable to top American systems. Graphistry researchers suggested the model may result from "illegal distillation" — a technique to replicate advanced AI models — of OpenAI and Anthropic tools.

Unlike their American counterparts, these Chinese models are "open-weight" systems that users can download and modify directly, including removing safety guardrails. Isaac Evans, Semgrep founder, called GLM-5.2's capabilities "a massive leap" and estimated that foreign adversary-linked hacking groups "are likely experimenting now."

Shrinking timeline

House Homeland Security Chair Andrew Garbarino stated that Beijing "is just months, if not weeks, away from achieving frontier AI capabilities comparable to those of the United States."

Margaret Cunningham, vice president of security and AI strategy at Darktrace, noted that open-weight models "make powerful capabilities more accessible and more difficult to govern through centralized controls."

Pearl framed the competition in stark terms: "The U.S. and China are truly in an arms race to develop and diffuse the most advanced AI capabilities. Like the Cold War, both the U.S. and China see this as a struggle with existential implications."

These details were first reported by Politico.