A Chinese tech executive stood before a cybersecurity conference in Beijing and declared that his company's AI bug-finding tool now rivals the most advanced American system, one the U.S. government has spent weeks trying to keep under lock and key. Zhou Hongyi, chief executive of 360 Security Technology, told attendees that his firm's tool, called Tulongfeng, is comparable to Anthropic's Mythos model, the very system Washington deemed too dangerous for broad release.
The timing could hardly be worse for American interests. While the Trump administration works to restrict public access to the most capable AI systems built by U.S. companies, Chinese competitors are racing to match them, and openly bragging about it.
Zhou's remarks, reported by the Wall Street Journal, came with a blunt declaration of intent:
"This kind of powerful weapon that can alter the landscape of cyberwarfare can't remain solely in American hands."
Zhou is not a fringe figure. His company, 360 Security Technology, is one of China's major cybersecurity players. And he is not alone. Another Beijing-based firm, Z.ai, also known as Zhipu AI, has reportedly narrowed the gap with American competitors in AI-powered vulnerability detection. Lior Div, CEO of cybersecurity firm 7AI, said China was "making sure that the gap becomes smaller and smaller over time."
That gap is closing in part because the U.S. government is hobbling its own champions. Earlier this year, Anthropic warned that its Mythos model was so adept at finding software flaws that it could be weaponized by malicious hackers and threaten critical computer networks. Fox News reported that Mythos discovered approximately 2,000 unknown software vulnerabilities in just seven weeks of testing, a pace that made it too dangerous for public release, in Anthropic's own judgment.
The concern is real. But the question is whether Washington's response is making things better or worse.
The chain of events moves fast. David Sacks, who co-leads Trump's council of technology and science advisers, described the origin of the alarm on a recent podcast:
"Dario came to Washington a few months ago, back in April, and basically said that he had created a cyber weapon called Mythos. And he spiked the cortisol level, got everyone really worried. And there was some truth to it in terms of the sense that this model had advanced cyber capabilities."
Anthropic CEO Dario Amodei's visit set off a cascade. Earlier in June, Trump signed an executive order establishing a framework for the federal government to vet national security risks of the most advanced AI systems for up to 30 days before their public release. Developer participation was described as voluntary, and the framework has not yet been fully developed.
Then things escalated. AP News reported that the Trump administration imposed export controls to prevent foreign nationals from accessing Anthropic's latest models, Fable 5 and Mythos 5. The directive arrived on a Friday afternoon, with no specific national security concerns cited, forcing Anthropic to pull both models offline just days after Fable 5's public launch.
The scope of the disruption was significant. Breitbart reported that the ban on foreign nationals forced Anthropic to disable both models for all customers, affecting hundreds of millions of users. The government's concern reportedly centered on a discovered method of "jailbreaking" Fable 5 that could potentially aid hacking activities.
Anthropic pushed back publicly. "We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people," the company stated. In a separate statement, Anthropic said it believes "the government should have the ability to block unsafe deployments, as part of a statutory process that is transparent, fair, clear, and grounded in technical facts. This action does not adhere to those principles."
The administration's pattern of engagement with China's broader strategic ambitions extends well beyond AI. Concerns about alleged efforts to sabotage U.S. AI infrastructure to benefit Beijing have surfaced in recent months, adding another dimension to the competition.
The Commerce Department effectively banned Mythos 5 about two weeks before the administration reversed course. On Friday, the government approved a limited redeployment, Anthropic confirmed that Mythos 5 was enabled to be "redeployed to a small group of cyber defenders and infrastructure providers."
Newsmax reported that the administration is expected to lift restrictions on Fable 5 as soon as the coming week. But for now, one of Anthropic's flagship consumer models remains offline, pulled from the hands of American businesses and researchers who had been using it freely.
The White House said Friday that it "continues to collaborate with frontier AI labs on addressing the challenges of scaling the fast-growing technology." That language is polite. The reality is messier.
Anthropic is not the only San Francisco AI company feeling the squeeze. On the same Friday, the Trump administration asked OpenAI to restrict the release of its newest model, GPT-5.6 Sol. OpenAI complied. Only about 20 Trump-approved customers can access it.
OpenAI stated that Sol "is better at helping people find and fix vulnerabilities" than it is at carrying out cyberattacks. But the company acknowledged uncertainty, saying: "That uncertainty, along with the model's broader step change in capabilities, is why we are pairing the model's increased capabilities with stronger safeguards and a phased release."
OpenAI also signaled discomfort with the arrangement. "We don't believe this kind of government access process should become the long-term default," the company said, adding that it sees a "path to broader availability in the coming weeks."
The pattern is clear: the government is picking winners and losers on a case-by-case basis, with no published criteria for who gets access and who does not.
Here is where the picture gets uncomfortable. While Washington restricts American companies from deploying their most advanced AI tools broadly, the administration has reportedly continued to allow AI chip exports to China. Saif Khan, a think tank fellow, called this combination a "gift" to Xi Jinping, letting Beijing keep importing the hardware it needs to build competitive AI systems while American firms are barred from releasing their own software to their own customers.
Niels Provos, a researcher who led security teams at Google and Stripe, offered a sharper critique of the downstream effects:
"It is incentivizing companies across the globe to use cheaper but very capable Chinese open-weight models, while at the same time undermining the US AI industry. I don't understand it."
That dynamic deserves attention. Chinese open-source AI models, available freely to anyone, anywhere, become more attractive every time Washington restricts access to American alternatives. Microsoft is reportedly considering offering Chinese AI models on its platforms. If U.S. companies cannot release their tools, the global market does not simply wait. It turns to the competition.
China's broader strategy of acquiring and replicating American technology is well documented. Beijing has built sprawling military infrastructure aimed at matching U.S. capabilities, and its approach to AI follows the same playbook: close the gap by any means available.
Jacob Helberg, undersecretary of state for economic affairs, told the Wall Street Journal that the administration is aware of the challenge. "Our administration is very much focused on Chinese open-source models," he said. "It's something that we're tracking very closely."
Tracking is not the same as solving. And the administration's approach has drawn criticism from both sides of the aisle. Rep. Lori Trahan, a Massachusetts Democrat who co-authored a bipartisan AI regulation bill, issued a pointed statement:
"The Trump administration is deciding company by company who gets access to the newest AI model. No law. No process. No oversight. Just appointees in Washington deciding who's in and who's out."
Trahan's critique, whatever its partisan motivation, identifies a real structural problem. The executive order Trump signed in June established a voluntary framework that has not yet been fully developed. In the meantime, decisions about which models can be released, to whom, and under what conditions are being made ad hoc, by government officials, behind closed doors, with no published standards.
The concern about AI-powered cyberweapons is legitimate. Mythos's ability to find 2,000 unknown vulnerabilities in seven weeks is a genuine national security consideration. But restricting American innovation while Chinese competitors openly announce they have matched it is not a strategy. It is a contradiction.
The pattern of Chinese institutional penetration into American systems, from university programs designed with CCP input to alleged infrastructure sabotage, makes the stakes even higher. Every month that American AI companies operate under ad hoc restrictions while Beijing operates without them is a month the gap narrows further.
And China's military ambitions leave no room for complacency. Beijing has been replicating U.S. military hardware at desert test sites. Its approach to AI cybersecurity tools follows the same logic: copy, match, surpass.
The administration faces a genuine dilemma. AI models capable of finding thousands of software vulnerabilities in weeks are powerful tools, for defense and for attack. Releasing them without safeguards could endanger critical infrastructure. But locking them away while Chinese competitors build equivalent systems in the open accomplishes something worse: it surrenders the field.
The executive order's 30-day vetting window could become a reasonable framework, if it is actually built out, made transparent, and applied consistently. Right now, it is a skeleton with no muscle. Companies are complying not because the rules are clear, but because the government asked and they felt they had no choice.
OpenAI said the quiet part out loud: this process should not become the long-term default. If it does, the United States will have created a system where its own AI industry operates under government-imposed handicaps while its chief rival operates freely, and closes the gap one conference presentation at a time.
You do not win a technology race by grounding your own planes.
