Trump’s Ai Flip-flopping Could Be A Gift To China

The Trump administration in recent weeks has meddled with the rollout of advanced artificial intelligence models developed by America’s top AI companies.

China, meanwhile, has announced a slate of new AI tools it says rival the United States’ most sophisticated technology at a fraction of the cost.

Some security experts worry that the White House’s efforts to restrict the U.S. tools from wider access — and dithering between promises of light-touch intervention and more invasive regulatory action — may have given China a window of opportunity in the race to develop AI tools that could be used as cyber weapons.

“If the administration is honest about wanting the United States to beat China in this race, then this is about the dumbest thing they could possibly do,” warned Alex Stamos, chief product officer at the AI security firm Corridor, during a briefing last week hosted by the Center for Democracy and Technology.

The U.S. has so far led the global development of frontier AI models with hacking capabilities. American AI giant Anthropic boasted in April that its newest Claude Mythos model can sniff out security flaws in software and operating systems better than most human minds. People who have studied OpenAI’s latest models call them similarly adept at finding vulnerabilities.

Earlier this month, the Trump administration placed export controls on two of Anthropic’s most advanced AI models over concerns that its security guardrails were not strong enough. And last week, OpenAI limited the release of its most powerful model yet, GPT-5.6, after the White House asked the company to restrict access.

The Trump administration fully lifted export controls on Anthropic’s newest models late Tuesday after weeks of negotiations, restoring access for companies and allied governments that have used the tools to fortify their networks against cyberattacks.

The merry-go-round of AI policies comes as American tech companies are navigating the contours of an early June executive order from President Donald Trump that asks AI developers to voluntarily submit their models to the federal government 30 days before public release for national security vetting. While that order rejected the idea of mandatory government controls, the administration’s export ban on Anthropic came less than two weeks later, following a hurried set of decisions that left supporters of both sides pointing fingers.

Matt Pearl, former director of emerging technologies at the National Security Council during the Biden administration, said the Trump administration is “engaged in an extremely difficult and complex balancing act” between protecting national security and preserving AI innovation.

“It’s critical for the administration to have a clear, transparent and fair process that applies to all model developers,” he said. “We can’t create an opaque bottleneck that prevents users in the U.S. and allies and partners from leveraging those models to protect national security.”

As American tech companies wait for clarity after repeated reformulations of AI policy, Chinese AI companies have announced a new wave of advanced AI systems — with capabilities they claim are comparable to those of their American competitors.

Chinese company 360 Security Technology unveiled two AI models last week that the firm said match the capabilities of Anthropic’s Mythos model. Reuters reported that the tools were designed to drastically improve vulnerability discovery and to automate incident response to cyberattacks.

A separate China-based company, Z.ai, has released its new model, GLM-5.2, which is around one-sixth of the cost of leading U.S. models. GLM-5.2’s bug-hunting capabilities were also found to be comparable to those of leading U.S. models, according to security assessments by the cyber firm Semgrep and the visual investigations platform Graphistry.

Researchers at Graphistry suggested in their findings that GLM-5.2’s capabilities may be the result of “an illegal distillation” — or tactic to steal and replicate advanced AI models — of powerful, American-made AI tools, including OpenAI’s GPT-5.5 and Anthropic’s Claude Opus 4.8.

Semgrep founder Isaac Evans said that while GLM-5.2 is “in a separate class from Mythos,” the model’s capabilities are still “a massive leap in capabilities amidst the other models.”

And unlike Anthropic's or OpenAI’s models, the ones from Chinese companies such as Z.ai are “open-weight” models that users can download and tweak directly. Those changes can include removing built-in safety guardrails that prevent hackers from weaponizing the model for cyberattacks.

“Open-weight models add another dimension because they make powerful capabilities more accessible and more difficult to govern through centralized controls,” said Margaret Cunningham, vice president of security and AI strategy at cyber firm Darktrace. Because of this, Evans estimated that foreign adversary-linked hacking groups “are likely experimenting now,” and cybercriminal groups will soon follow.

Spokespeople for Z.ai and 360 Security Technology did not respond to a request for comment.

Recent estimates suggested that Washington has a six- to 12-month runway before Beijing catches up to American AI capabilities.

But security experts and Capitol Hill cyber hawks fear that timeline may already be shrinking, and the limited release of American-made cyber-capable models is making it even harder for cyber defenders to prepare their networks for a future barrage of AI-powered cyberattacks.

House Homeland Security Chair Andrew Garbarino (R-N.Y.) said in a statement that Beijing “is just months, if not weeks, away from achieving frontier AI capabilities comparable to those of the United States.”

“The U.S. and China are truly in an arms race to develop and diffuse the most advanced AI capabilities,” Pearl said. “Like the Cold War, both the U.S. and China see this as a struggle with existential implications.”