US-China AI competition shifts as regulation clashes with China's cost-efficient model strategy

What would multilateral 'AI arms control' look like?

Given the competition, it's debatable whether a US-China safety deal is even possible The writer is author of 'Chip War' With the strictly limited release of OpenAI's GPT 5.6 model last week, it isn't only Anthropic that's bearing the brunt of the Trump administration's newfound zeal for regulating AI. Sam Altman, OpenAI chief executive, announced his company's latest model while noting delicately that "this isn't quite the process that we think is optimal". Yet America's AI regulation debate isn't limited to the White House and leading AI labs. It is being shaped by competition with Beijing. Could China and the US agree to rules that would address the risks in advanced AI systems while in the middle of the race? The frontier model contest changes nearly daily but Chinese companies continue to progress. The latest example is Z.ai's newly released GLM 5.2 model, whose capabilities led David Sacks, co-chair of President Donald Trump's Council of Advisors on Science and Technology, to declare: "We now have a Chinese open-weight model that is as good as the currently available models from OpenAI and Anthropic." In fact, model quality is growing harder to compare as the industry questions the efficacy of standard benchmarks such as mathematical problem completion. These benchmarks can be gamed if the model is trained on the specific problems in advance. What's more, as OpenAI researcher Noam Brown recently noted, the right definition of capability is not only what problems a model can solve but how quickly and at what cost. For complex, long-horizon tasks -- the ones Anthropic and OpenAI are focused on -- US companies may well have a longer lead than simple benchmarks suggest. But there's no doubt Chinese models are improving. Before Trump, the Biden administration had hoped its AI chip restrictions on Beijing might give the US such a commanding lead in AI that Washington could dictate terms. This compute advantage has kept US companies ahead when measured by model capability or revenue, but it has not prevented Chinese competitors from releasing their own high-quality models. If Trump tightly regulates US companies he will kneecap American AI leaders without fully addressing global security concerns. There is one alternative to unilateral regulation: a deal between China and the US to address AI safety. The Trump administration appears keen for talks with Beijing. The two sides reportedly discussed the topic before the May Xi-Trump summit, though Beijing has not agreed to formal negotiations. So it is worth asking what a multilateral "AI arms control" regime could look like. To start, we need to take cyber security -- the reason that the White House is restricting foreign access to Anthropic's Fable model -- completely off the table. It's nearly impossible to imagine the two countries agreeing not to use AI for cyber operations because it's already happening so widely. Both Chinese hackers and the US National Security Agency use leading Anthropic models for this purpose, according to reports. This isn't surprising: any model good at computer programming will be good at identifying cyber vulnerabilities. A second regulatory priority is limiting the ability of AI to enable production of dangerous pathogens. It's not impossible to imagine that China and the US could develop shared principles for assessing a model's biosecurity risk. However, China's record of biosecurity transparency is not reassuring. Nor are the cold war arms control parallels. When the US and Soviet Union signed the Biological Weapons Convention in 1972, the US had already begun dismantling its bioweapons programme. But the Soviets, convinced the US was lying, expanded work on anthrax and smallpox. Could AI controls extend into the sphere of weaponry? China's pre-existing support for talks on banning autonomous weapons may sound reasonable. But will any major power renounce use of autonomous missiles or automatic missile-defence systems? An even bigger issue is assessing compliance. If a country promised not to use AI for a specific purpose, how could this be proved? It was hard enough to count missile silos during the cold war, even though these were visible from space. Neither China nor the US will give the other access to sensitive source code. "Trust but verify" was Ronald Reagan's wise approach to US-Soviet talks in the 1980s. So long as we distrust and cannot verify, we should not expect a significant deal. Both Washington and Beijing are likely to conclude that the priority is to keep racing instead.

Opinion | The Real A.I. Race Isn't America vs. China

Mr. Wang is a research fellow at Stanford's Hoover Institution. Mr. Gewirtz served in senior China policy roles in the Biden administration's National Security Council and State Department. A globally recognizable tech executive, high-spirited from preparing for a public offering, offers imprudent remarks criticizing the government. The state strikes back harder than anyone expects. Overnight, the bargain between a skyrocketing sector of the economy and the government is shattered. If you think this story could be about Anthropic, you're only half right. In 2020, the Alibaba co-founder Jack Ma found himself in the doghouse after he publicly rebuked Chinese regulators. Citing regulatory concerns, the authorities canceled the public offering of Ant Group, another company Mr. Ma helped found, and subsequently unleashed a regulatory storm that left few Chinese tech companies untouched. The U.S. government is skating close to its own Jack Ma moment, when a government wounds a tech leader seemingly out of spite. Self-destructive American actions, not Chinese competition, may be the most significant threat to the evolution of A.I. for years to come, long after the government and Anthropic resolve their current dispute. On June 9, Anthropic released its model Fable 5, an adapted version of its powerful Mythos model, which has incredible capacity to find vulnerabilities in software. Anthropic's chief executive, Dario Amodei, even said that companies that used Mythos had called it a "superweapon." Three days later, the U.S. government issued an export control directive blocking use of Fable 5 by foreigners and noncitizens -- including some of Anthropic's own employees -- which prompted Anthropic to disable all access to the model. A tangle of explanations behind this directive have been reported, including the risk of "jailbreak" (when a model bypasses built-in safety guardrails) and the risk of access by foreign adversaries. On Friday, the government permitted Anthropic to restore some users' access to a version of Mythos, though negotiations are still underway about Fable 5. Over the past decade, the U.S. government has used export controls to deal sometimes crippling blows to Chinese technology champions. The action against Anthropic upended this logic by turning this policy instrument against American companies, purportedly to exert the U.S. government's grip over increasingly slippery A.I. models. The challenges of regulating A.I. at the frontier have now reportedly moved the government to ask Anthropic's chief competitor, OpenAI, to limit the users for its own next model. We often think of A.I. as a race between the United States and China. Instead we are seeing the emergence of an even more acute form of competition, between the public power of governments and the private power of ambitious companies. Both countries are struggling to determine whether their frontier A.I. companies are national champions or national security threats. A.I. labs in both countries are also starting to realize how much their operations depend on the state's sufferance. The U.S. government needs to strike a better balance between ambition and control, lest it irrevocably damage its relationship with these companies and America's long-term technological edge. The second Trump administration has careened from extreme to extreme on A.I. policy. It came into office de-emphasizing some A.I. safety concerns, downplaying harms to workers and preaching the virtues of a hands-off approach. Its stance shifted in March: The Pentagon designated Anthropic a supply chain risk after the company protested the use of its A.I. models in autonomous weapons and domestic mass surveillance. Reports about Mythos's capabilities the following month appear to have shocked the administration into taking safety more seriously. The confrontation around Fable 5 has been the latest turn in a dance between a company that proclaims a desire for safeguards and an administration that wants unfettered control over these models. The Chinese government, meanwhile, has been overseeing its A.I. labs for considerably longer. It requires security assessments and tests models for their ability to anticipate Beijing's sundry political sensitivities. The state has blocked Chinese companies from purchasing advanced Nvidia chips in an effort to favor the domestic chip industry, even after the Trump administration cleared these chips for sale to China. It has reportedly restricted overseas travel by certain A.I. researchers and told two of the founders of Manus, a popular A.I. agent that acts as a digital assistant, not to leave China while the government reviewed their company's acquisition by Meta. The authorities soon after ordered the unwinding of the deal. Unlike in the United States, in China no one questions who will ultimately win the struggle between state and corporate power. But the effects of the regulatory storm Beijing unleashed in 2020 remain evident. Alibaba's stock has lost around two-thirds of its value since Mr. Ma's speech and Ant Group has still not been able to go public. Venture capital funding in China collapsed and is only now rebounding. These days, nearly any conversation with Chinese A.I. labs includes bitter complaints about government overreach. The United States, with its system of legal protections and its deep capital markets pouring money into A.I., is not likely to suffer to the same degree. The companies and government officials are likely to reach an understanding. But Washington and American A.I. labs need to treat each other with greater sincerity and seriousness going forward. First, the heads of A.I. labs need to stop their doom trolling. It makes little sense to make panicked claims about the destructive potential of A.I. without a plan to work with the government to address those risks. The current administration makes such cooperation particularly difficult because of its demonstrated willingness to punish companies to assert dominance or satisfy a political constituency. But the labs still must take more care to address A.I.'s transformative impacts. Second, the U.S. government needs to realize that the stakes of A.I. are far too high to allow a breakdown of trust. Despite a recent executive order that established a voluntary review program of highly advanced models, Defense Secretary Pete Hegseth has seemingly delighted in celebrating Anthropic's troubles with the government. The administration has imposed major restrictions on Anthropic without offering sufficient explanation. Even if the government is nettled by the attitude of A.I. chiefs, it is irresponsible to treat them with contempt. American allies are wondering if they can rely on American A.I. models, and talented foreign researchers in American labs are reconsidering their career plans -- neither of which serves American interests. Third, America's government and its A.I. labs need to update their understanding of Chinese A.I. Beijing's pronouncements on A.I. are distinctly less apocalyptic than Anthropic's; Chinese firms are less able to invest in data centers (for lack of American chips) and more interested in applying A.I. to physical technologies. By most appearances, China is pursuing a substantially different vision of the A.I. future. Frictions within China may present an opportunity for the United States to put further distance between Chinese entrepreneurs and the overbearing Chinese state -- but that would require the administration to drop policies that seem hostile toward foreign talent, including Chinese citizens. The A.I. chaos of recent weeks is self-defeating. For the United States to win the A.I. future, it needs to do better at avoiding the missteps of Beijing. Dan Wang is a research fellow at Stanford's Hoover Institution and the author of "Breakneck: China's Quest to Engineer the Future." Julian Gewirtz, a senior research scholar at Columbia University, served as senior director for China and Taiwan Affairs at the National Security Council and deputy China coordinator at the State Department. The Times is committed to publishing a diversity of letters to the editor. We'd like to hear what you think about this or any of our articles. Here are some tips. And here's our email: [email protected]. Follow the New York Times Opinion section on Facebook, Instagram, TikTok, Bluesky, WhatsApp and Threads.

Opinion | Is the future of AI open?

The leading artificial intelligence labs have a strong interest in keeping open-source competitors at bay, and Trump administration officials need to be wary of those motives as they exert more control over models. OpenAI CEO Sam Altman is reportedly in talks to give the U.S. government a 5 percent stake of his company to clear political obstacles. Anthropic convinced Commerce Secretary Howard Lutnick to drop export controls on the Mythos and Fable models this week by imposing stricter limits on cybersecurity prompts and "proactively" addressing security concerns. Both firms run highly lucrative, closed models. Only their engineers know the mathematical formulas, known as weights, used to answer queries. Both companies are preparing initial public offerings at blockbuster valuations that assume this is the future of AI. It might not be. Nvidia, the chipmaker, is developing its own open-weight models as loss leaders. Anyone can see and adjust the weights as they see fit. CEO Jensen Huang's logic is that more companies building commodity AI into their products will drive continuing growth in chip sales. Palantir announced this week that it will run Nvidia's open Nemotron models inside classified, air-gapped government systems, because a customer can own the weights, inspect them and seal them off. Palantir CEO Alex Karp, hardly a dove on national security, argued that it's actually riskier for clients, governments or companies to hand their data to a third party -- even patriotic firms like the frontier labs -- than to secure it themselves. Up to now, China has been most closely associated with the open-source approach to AI. Chinese versions, by some accounts, trail American frontier models in cyber capabilities by only a few months. These models, hosted on servers beyond the reach of any export control, are already being downloaded not just by malicious actors but also firms trying to control spiraling costs. Nvidia's third-generation Nemotron models still trail China's best, but the gap is narrowing. The temptation for the U.S. government, perhaps egged on by the frontier labs, will be to reach for a regulatory hammer. If the feds cannot control the release of a Chinese model, they could try to criminalize its use: Bar U.S. platforms from hosting the weights, and punish U.S. businesses for building products around them -- in the name of national security. That temptation will be worth resisting, as it would hurt no one but American start-ups and enterprises adopting these models precisely because they are competitive, cost-effective and customizable. Regulation would limit U.S. adoption, putting companies at a disadvantage against foreign competitors who would remain free to adopt cheap Chinese alternatives. China's labs may be building their models partly by training on the outputs of American ones. Distillation is the technical term. This is a very legitimate grievance. But, economically, they are rowing in the right direction: Open-source models are driving the price of frontier-grade AI toward zero and putting it in everyone's hands. The right answer is more competition, not picking winners and losers.

China's AI progress strains U.S. alliance pitch

Why it matters: Chinese models don't have to beat OpenAI or Anthropic to reshape the global AI order. They just have to be useful, available and widely adopted. Between the lines: Experts argue that two key things are kneecapping the U.S. government's desire to export American AI: * An erratic export controls strategy that involves making decisions about access to advanced models on the fly. * Not paying sufficient attention to China's efforts to spread its open-source AI models abroad while deploying AI at scale across manufacturing, health care and other industries domestically. Driving the news: The State Department this week expanded its Pax Silica initiative, an effort to build a U.S.-led AI and chip supply-chain bloc and reduce reliance on Chinese technology. * The move came as the AI industry continues to grapple with the fallout of the government's decision to put export controls on Anthropic's newest AI models. * Meanwhile, Chinese AI is closing the capability gap, as Axios' Sam Sabin has reported, and the models are much cheaper. What they're saying: "I think we're seeing another example of the Huawei strategy in the context of open-source AI models," Emily Weinstein, a former Commerce Department staffer now at the U.S.-China Economic and Security Review Commission, told Axios. * "China is able to offer not even just the models, but often the underlying or associated infrastructure at either no cost or significantly lower cost," said Weinstein during a panel discussion held by the Center for a New American Security this week. * Widespread adoption of Chinese AI in the Global South could create a "Huawei model on steroids" where countries are reliant on Chinese infrastructure that is not interoperable with the U.S., Weinstein said. Following the Anthropic decision, "the entire industry is kind of frozen in place, waiting for something that seems kind of more coherent," CNAS' Daniel Remler, a former State Department technology adviser, said during the panel. * "That's concerning when the Chinese are trying to move as fast as possible," he said. * "You're seeing many more calls now for AI sovereignty. I think it will mean that much of the rest of the world will likely, at least on the margin, prefer Chinese open-weight models" over U.S. frontier models, said Saif Khan, former counselor for critical and emerging technologies to the Secretary of Commerce. Yes, but: The Trump administration is trying to counter that trend by getting allies on board with American AI instead. * This week, undersecretary of state for economic affairs Jacob Helberg announced that 35 countries signed the "Declaration on AI Opportunity" as part of Pax Silica. * Helberg also criticized digital sovereignty -- the push for homegrown AI infrastructure -- as "backward and counterproductive" and "synchronized mediocrity." Reality check: Some U.S. partners are walking a tightrope, embracing the administration's vision while also pursuing greater technological sovereignty. * The European Union notably touted the importance of digital sovereignty following the Anthropic decision. * In an interview with Axios while in Washington for government meetings and the Pax Silica event, Omran Sharaf, the United Arab Emirates' assistant foreign minister for advanced science and technology, said the country's goal is achieving "strategic autonomy through international collaboration with trusted partners." * "For the U.S. to be able also to maintain its global position, its economic weight and posture, even national security, it's important for the U.S. to be able to have their technology deployed around the world," Sharaf said. * Sharaf added that it's "too early to tell" what export control decisions like the Anthropic directive will mean for that goal. The bottom line: The next challenge for the U.S. isn't just staying out front on frontier AI. It's convincing the rest of the world to keep building on American AI.

In AI race vs. U.S., China eyes victory in lower prices and broader appeal

Workers assemble and debug AI robots in Beijing. (Yi Haifei/China News Service/VCG/Getty Images) SINGAPORE -- In the competition for global dominance in artificial intelligence, Chinese AI leaders increasingly see themselves running a different race than their American counterparts, focused not on developing the most sophisticated capabilities, but on offering "good enough" models at cheaper prices that will proliferate more widely. In Shanghai and Hangzhou, China's AI hubs, many no longer see the "frontier" as achieving technical benchmarks, said Cherie Shi, global business manager for the Chinese AI company MiniMax, but rather "how many people in the real world are actually using our models every day." Over the past year, Chinese AI firms have gained significant ground in promoting adoption of their models by companies, governments and individual users in places from Southeast Asia to the Persian Gulf region. A government-backed AI initiative in Singapore recently chose to build its capabilities on Alibaba's Qwen. A Malaysian property conglomerate is embarking on a $20 billion "smart city" project featuring an AI research center backed by Chinese tech firms. Saudi Arabia is partnering with ByteDance and Huawei to apply AI in urban infrastructure. U.S. leaders have pitched the battle for AI preeminence as a contest over commanding the future. "Whoever leads that is going to really lead the world to a large extent. That's how big it is," President Donald Trump said earlier this month. The United States, he told reporters, is "leading China by a lot." But analysts and investors in the AI sector say that lead may be misleading -- and short-lived. "Top models may converge near the frontier, but outcomes increasingly depend on prompt quality, tools and domain integration," the JPMorganChase Center for Geopolitics said in a May report. Chinese AI labs, it added, are producing models primed to "win at adoption." After initially scrambling to get employees to incorporate AI into their work, companies are now reeling from the skyrocketing costs of AI use and seeking ways to stretch the utilization of "tokens," which are the basic units of AI data consumption. This has opened opportunities for Chinese AI firms, which offer cheaper rates for tokens and build on models that operate more efficiently, those in the industry said. Many enterprises need "a fraction of the capability" offered by leading U.S. firms like Anthropic and Open AI, said a salesperson at a leading Chinese AI firm, who spoke on the condition of anonymity because he was not authorized to speak to the press. "If we can provide 80 percent of the value at a much lower cost, that's enough. We'll take that market share." Gunja Gargeshwari, chief revenue officer for Bright Data, a Tel Aviv-headquartered tech firm providing web data for AI, said many of his company's clients are experimenting with Chinese AI. "It's undeniable," Gargeshwari said, "the cost efficiency, the token efficiency. It's amazing." Even as the U.S. government amps up its concerns over the national security risks of Chinese AI, Chinese firms "are getting a seat at the table" in many meeting rooms -- a change from just a year ago, Gargeshwari said. "Companies are not biased against China," he added. "They are very open to using the right solution for their enterprises. ... So the future is bright from that perspective for Chinese makers." In mid-June, Zixuan Li, an executive at one of the most promising Chinese artificial intelligence start-ups took to a stage at Singapore's glitzy Marina Bay Sands for a conference called SuperAI. Top U.S. companies are selling Rolls-Royces, while top Chinese firms are selling Mercedes, said Li, the director of product at Z.ai, known in China as Zhipu AI. The standing-room-only audience grasped his point: American AI is better -- for now. But Chinese products are not far behind in capability, more affordable and, eventually, likely to be far more widely sold. Restricted by the U.S. Commerce Department for alleged ties to China's military, Zhipu recently opened offices in Singapore and Dubai, and it is picking up government contracts across Asia, salespeople familiar with the deals said. Zhipu has objected to being on the Commerce Department entity list, which limits designated entities from doing business in the U.S. based on national security concerns. In a blog post last year, Open AI singled out Zhipu for its aggressive overseas expansion, writing that it was working to "lock Chinese systems and standards into emerging markets before U.S. or European rivals can." The company's marquee product is a model called GLM, which is popular among coders. As of March, the company said it had more than 4 million registered users in 218 countries. Chief scientist Tang Jie suggested GLM could match the capabilities of Anthropic's latest model, known as Fable 5, before the end of the year. The Chinese government has poured hundreds of billions into companies like Zhipu in recent years and helped them broker deals with countries along what it calls the "Digital Silk Road" to fuel the outward push on AI. The geopolitical goal is for China to "challenge the West's long-standing dominance over rule-setting" in digital technology, according to researchers at the China Academy of Macroeconomic Research and the Economic Information Network, two state-affiliated think tanks. This push is also ramping up at a time when U.S. policy toward AI diffusion seems increasingly convoluted, analysts say. In mid-June, the Trump administration blocked foreign nationals from using Anthropic's most powerful artificial intelligence models, sending stocks of Chinese AI companies, including Zhipu, soaring as much as 48 percent. Before this, Anthropic had imposed its own restrictions on how Claude, its signature large language model, could be used for military purposes, clashing publicly with the Trump administration. Users are "reasonably concerned" that access to American AI products is unreliable, said Martijn Rasser, a vice president at the Special Competitive Studies Project (SCSP), a nonpartisan initiative on tech leadership based in D.C. "It's not an attractive feature," Rasser said. At a forum in Paris last week, Alibaba's chairman, Joseph Tsai, referenced the curbs on Anthropic and called on Europe to consider China's AI offerings. "The problem is right now all your eggs are in one basket," Tsai said. "Why not choose a second basket and diversify your eggs?" White House spokesperson Liz Huston did not respond to queries on the Trump administration's clashes with Anthropic, saying only that the administration has adopted a "whole-of-government effort" to pursue AI dominance. Apart from affordability, the other major draw of Chinese AI products is that most are open-source, meaning they can be downloaded and used locally instead of over cloud-based services. This appeals to governments that want to deploy AI from local data centers that they control, said Shi at MiniMax. It also attracts entrepreneurs and start-ups handling sensitive information, such as banking details or health metrics, and users who are wary of being beholden to AI companies that can increase their prices at any point. Aleksander Mordvinov, a Russian living in Thailand, said his entire AI banking start-up was built on Qwen because he wanted data to be controlled locally. "All companies who have cloud systems sometimes have accidents where the data leaks. I can't have that," Mordvinov said. "This guy," he said, grinning as he picked up a plushie version of Qwen's capybara mascot, "this guy is what works for me." By focusing on achieving technological advances at the frontier of AI development, and neglecting the commercial application and spread of their products known as diffusion, American AI companies risk ceding "soft power and economic clout" to China that could ultimately give Beijing greater influence in shaping the global AI sector, Rasser said. If China can capture greater market share, it can also pour those profits back into advancing its tech, as it has done with solar equipment, rare earth processing, electric vehicles and other industries, said William "Chip" Usher, another SCSP analyst. But others in Washington and in Silicon Valley think it is still more worthwhile to focus on the cutting edge. Their argument, Usher said, is that there are capabilities beyond the frontier that will be transformative once achieved. Bo Bai, a Chinese-born U.S. citizen, said his Singapore-based fintech company MetaComp initially started building its AI capabilities on open-source Chinese models. But in October, Anthropic released a powerful new feature called "Skills," which allows users to teach Claude repeatable workflows. The update stunned Bo and his engineers. "We threw away everything," he said. "And then we restarted using Anthropic's Claude system."

US-China AI race: How America's own restrictions are helping China close the gap

The decades-long presumption informing US AI policy strategy has been straight forward: stay ahead, lock it down, and the lead will last. Today, that presumption seems frighteningly vulnerable - and ironically, the US might be doing more to close the gap than China ever could on its own. The crisis erupted this month in two parts. First, the Trump Administration demanded Anthropic shut off access to its top-of-the-line AI technologies, Fable 5 and Mythos 5, to anyone not a US citizen - including within US borders. Almost immediately, Chinese AI systems were able to perform at the same level as Anthropic's Mythos when used in cybersecurity applications - according to scientists, this is expected to turn the tables completely on the technology race around the world. The timing was no accident; it was a warning of what critics of strict AI export policies have long foreseen. Also read: Anthropic, OpenAI model restriction: AI juggernaut slowing down? China's Mythos moment The recently released GLM-5.2, created by China-based Zhipu AI, also referred to as Z.ai, became one of the ten most popular AI models worldwide. During some benchmarking evaluations performed by the cybersecurity company Semgrep, GLM-5.2 surpassed Anthropic's Claude Opus 4.8. With more instructions, both Opus 4.8 and GLM-5.2 are capable of performing as Mythos in finding bugs. The Chinese cybersecurity community did not have to wait for the results of the benchmarks to express its plans. During the ISC.AI 2026 conference on cybersecurity held in Beijing, the founder of 360 Security, Zhou Hongyi, introduced two AI security tools under the name "Yitian Tulong," which comes from a classical Chinese novel about martial arts. The Tulongfeng system was developed to find software vulnerabilities automatically. Zhou even referred to the system as "Mythos made in China." His statement was very clear and direct: "Such kind of powerful weapon that can change the face of cyber offense and defense must not be owned by others." The own goal Also read: Motorola Edge 70 Pro Plus vs OnePlus 15R: Which phone has faster performance? The justification provided by the US government for restricting the use of Mythos and Fable was seemingly unimpeachable: powerful cybersecurity AI in the wrong hands can be dangerous to national security. In fact, when announcing the launch of Fable 5, Anthropic itself admitted that the capabilities in cybersecurity provided by their model "could be misused to cause serious harm" if not used properly. But that's where the logic falls apart, as restricting the availability of something does not render it non-existent, merely drives people elsewhere to seek it out. According to security researcher Niels Provos, an individual who has led security teams at Google and Stripe, such measures will only "incentivize companies across the globe to use cheaper but very capable Chinese open-weight models, while at the same time undermining the US AI industry". In his words, he simply "doesn't understand the logic behind it". Apart from harming market share, the export directive aimed at foreign nationals will likely encourage Chinese-born AI researchers who are currently working for institutions like Anthropic and OpenAI to move back to China. Indeed, the political factor has added a layer of complications to the situation. In February, Donald Trump directed all US federal agencies not to use Anthropic's AI models since the firm refused to comply with the Pentagon's contract provisions that demanded all AI purchased by the Department of Defense be usable "for any lawful purpose" - Anthropic tried to receive exceptions concerning autonomous weapon systems and extensive domestic surveillance. The Pentagon then branded Anthropic as "supply chain risk." Initially just a policy dispute, this has now transformed into a punishment of the AI laboratory on an industrial scale according to external observers. This does not mean China has surpassed the US in AI capabilities. Although GLM-5.2 is comparable with the best US AI models in terms of cybersecurity, it still falls behind the products of Anthropic and OpenAI in other metrics. The gap still exists but is closing quicker than expected because of the unstable policy environment in Washington. The US set out to protect its most powerful AI from adversaries. What it's actually done is hand those adversaries a marketing opportunity, accelerate their development timeline, and telegraph to the rest of the world that American AI is unreliable infrastructure. That's not a security strategy. That's a self-inflicted wound.