I’m Vincent Chow, senior AI reporter with the SCMP's technology desk. I'm writing to you this week from bustling New York City, fresh off one of its best summers ever with the New York Knicks winning their first NBA championship in 53 years and a highly charismatic new mayor scoring high marks among most residents in his first nine months in office.

But with the NBA finals and the FIFA World Cup now firmly in the rear view mirror, the city is also now hosting another major event: in a courtroom in Brooklyn. That would be the long-awaited trial of Chinese tech giant Huawei Technologies, which is getting its day in court eight years after the US government first brought criminal charges against the company. 

As one of the first major Chinese tech companies to land in the US government’s crosshairs, Huawei’s treatment by the US government has always had, rightly or wrongly, additional significance. In the years since it was first charged in 2019, the US government has gone after other Chinese tech companies and sought to pin similar labels of “Chinese government-controlled company” on them, threatening them with sanctions and export controls the same way Huawei has been treated.  

But like Huawei, many of these Chinese tech firms continue to flourish despite the US government’s efforts to thwart their development over what Washington sees as national security concerns. That continues to be the case in the AI era, with the US government increasingly prioritising AI as the most important front in US-China tech competition, while Chinese AI companies look to follow in Huawei’s lead and defy US restrictions.   

US keeps up AI offensive; China remains defiant

US keeps up AI offensive; China remains defiant

Accusations against Chinese tech companies of illicit activity are not new, especially from the US. Still, the last few days have seen a remarkable escalation in how Washington and Silicon Valley are calling out China’s AI industry, most notably on the controversial issue of AI model distillation — the alleged practice of Chinese companies “stealing” the outputs of US AI models to train their own.

On Tuesday, three US government agencies including the FBI and the NSA accused six Chinese companies of engaging in an industrial campaign to steal US model outputs, “likely with Chinese government awareness”. The report from leading US law enforcement and intelligence agencies was soon followed by a report from leading US AI company Anthropic, which claimed that its Chinese AI rivals including national champions DeepSeek and Moonshot covertly routed customer requests to its Claude models and passed those responses as their own. 

AI and legal experts say that the claims mark a new intensity in US criticism of China’s AI industry, given the unprecedented detail provided in the reports and the significance of the agencies behind them. 

What is notable is that Chinese AI companies accused of such malpractice have largely stayed mum about these allegations. Instead, they have left it to officials in Beijing to respond, with China’s Ministry of Commerce saying on Wednesday that Washington is using these allegations to unfairly suppress China’s AI industry. 

So what are Chinese companies focusing on instead? The answer is continuing to grow their businesses. Both Moonshot and DeepSeek, the two companies highlighted by Anthropic in its latest report, are working toward blockbuster IPOs that would secure them the funds needed to compete in the global AI race. 

Hangzhou firm DeepSeek is looking to list in Shanghai’s Nasdaq-like Star market with a valuation of around 500 billion yuan (US$74 billion) before investment. On Wednesday, my colleague Ben Jiang reported sources as saying that the Hangzhou firm has now hired underwriters including Citic Securities. The company is also in the midst of a massive hiring and infrastructure buildout spree, fresh off the release of a new model that once again reaffirms the start-up’s status as one of the most cost-effective operations around.

Moonshot, meanwhile, is also focused on raising money. My colleague Coco Feng reported on Thursday that the young start-up is exploring a dual listing in both Hong Kong and Shanghai to access more capital and exposure, in a sign of the Beijing-based company’s ambitions.

And what about Huawei itself, now the defendant in a high-profile trial where it could be subject to one of the biggest corporate fines in US history if found guilty? 

Earlier this week, just as its trial was getting started, the Shenzhen-based company announced a new smartphone powered by its first processor designed using the Tau Scaling Law the company announced in May. It also announced a new high-speed optical module designed to eliminate bottlenecks in AI infrastructure, emphasising its leading role in global technology standard-setting.

Through its extensive sanctions and legal regimes, Washington continues to drive global discourse surrounding Chinese tech companies. But what Huawei, and now other Chinese tech companies, are increasingly showing is that there is a viable path to flourishing despite US export sanctions. As the two country’s presidents meet in Washington later this month, the big question is the extent to which this dynamic proves to be an enduring one — or whether new actions by either side upend things entirely. 

Hot Seat

Hot Seat

In this new section of the CFT newsletter, we ask experts and industry leaders at the frontlines of US and Chinese AI development for their quick takes to the hot button issues of the day.

This week's guest is Joe Khawam, managing director of legal and AI policy at the Law Reform Institute in Washington, and a former attorney at the US Department of State. His current research focuses include model distillation and the legal ramifications of the controversial practice. 

1. Anthropic has accused multiple Chinese AI companies of “industrial scale” distillation campaigns to “steal” the outputs of its Claude models, a couple days after three US agencies including the FBI and the NSA released an advisory about Chinese companies’ alleged distillation campaigns. As a legal expert, what jumps out at you from these latest reports? 

Both the three-agency report and the Anthropic report have added a lot of details that weren't previously public. I think there are lots of questions that are going to keep lawyers busy here. There are details here that are potentially legally relevant, but whether they change the analysis under various US statutes, I'm not so sure yet, because it's going to take some time to do that analysis. 

On the US side, there were further details on efforts to extract Claude’s reasoning traces (the “chain of thought” reasoning tokens that AI models today use to “think” before generating their final answer). I think that strengthens the argument that [the US] can look at this from a trade secret theft perspective.

That said, most of the literature based on the information up until now had assumed that the strongest criminal theory would be under the Computer Fraud and Abuse Act (a US cybersecurity bill enacted in 1986 that criminalises IT-related fraud involving false identities and unauthorised access). I suspect that's still the case.

2. What is the significance of the timing of these reports, just weeks before a highly anticipated meeting between Chinese president Xi Jinping and US president Donald Trump, as well as AI-related talks between the two countries? 

I think the issue is certainly developing quite a bit ahead of the US-China talks coming up this month.

The argument I’ve made before was that strengthening their defences is something that the companies need to be doing. And the US government needs to be working with them to do that.

But that likely is going to be insufficient to deter continued adversarial distillation attacks because of the fact that, for every defence that's erected, a new technique can be developed to get around it. 

So in order to actually create deterrence here, there needs to be some sort of measure taken by the US government to create costs for those engaged in these types of distillation campaigns. That's where tools like entity listings and sanctions come in. 

But in an ideal world, the US government would engage in quiet diplomacy with the Chinese government to attempt to achieve the result they're seeking without having to resort to these types of measures.

3. What might the US government do to restrict its own companies, such as US cloud providers, from working with these Chinese companies, which are being accused of stealing from leading US AI companies? 

Generally, these types of cloud services are not considered an export under US export control law: they're considered to be a service. That's why in the past, the US Bureau of Industry and Security, or BIS, has issued advisory opinions essentially saying that these are services, not exports.

However, there have been recent reports that BIS is drafting a new AI diffusion rule for AI chips. As part of that rule, it may contain a provision regarding cloud services, so it's certainly possible that BIS may take a different position on that issue than they've taken in the past.

How precisely they would do that, I'm not sure. However, it certainly seems like an area that is very ripe for some sort of US government action because it’s a pretty big legal gap in US efforts to constrain the compute (computing power provided by semiconductor chips) provided to Chinese AI model developers.

Do US sanctions and scrutiny in fact make Chinese companies stronger by forcing them to be more innovative? Is China's AI industry only capable of replicating the successes of its US counterparts? Share your thoughts with me at vincent.chow@scmp.com. For those in New York City, do reach out if you're interested in chatting. 

The best of SCMP Tech