China Escalates AI Talent Curbs as Trump Floats Arms Sale to Beijing

On 28 September, Business Standard carried a Bloomberg report saying Beijing has broadened overseas travel restrictions on leading AI professionals at private firms to include their immediate family members. Spouses and children of certain AI and chip executives now need approval before going abroad, even for short trips.

The report cites people familiar with the matter. It says government agencies have begun telling affected individuals about the wider restrictions, among them startup founders and heads of strategically important AI companies. This is not an outright ban. But it adds a family-level layer to curbs that already covered the professionals themselves, and it is the clearest sign yet of how far Beijing will go to stop know-how from walking out.

Who is covered, and what is not clear

According to the Bloomberg account, the government is now targeting talent in the private sector specifically, rather than the state researchers, nuclear scientists and state-firm executives who have faced travel limits for years. Those covered include staff at firms such as Alibaba Group Holding Ltd. and DeepSeek, whose foreign travel has been restricted since earlier this year, Bloomberg News reported in May. That earlier reporting is background now. The new element is the family.

The report also says it is unclear whether all those people's families must follow the new strictures. The Ministry of Industry and Information Technology did not respond to a faxed request for comment. Two of the people said government departments have already notified several individuals that they need to clear family travel.

The timing sits alongside a separate set of rules that took effect on 15 September for enforcing exit bans in cases ranging from criminal investigations to potential violations of the country's industrial and technological security. Business Standard says Beijing plans to add relevant individuals to the list over time.

The report ties the broader anxiety to Meta Platforms Inc.'s $2 billion acquisition of Manus, which the Chinese government eventually ordered the US firm to unwind. It notes the travel plan is not necessarily linked to the Manus saga, but that guarding against technology leaks remains a policy goal.

That framing matters. China's top AI talent pool largely emerged after ChatGPT, mostly inside tech giants or private startups. A policy that reaches spouses and children forces engineers with global ambitions into an earlier choice: stay home, or go abroad before the constraints tighten. The Bloomberg piece puts it that way, and the logic is hard to argue with.

Humanoid robots and the export story Beijing wants told

While the talent rules tighten at home, China's export machine keeps producing numbers that Western capitals read as strategic. On 29 September, TechNode reported figures attributed to IDC showing China shipped more than 19,000 humanoid robots in the first half of 2026, or 77.9% of global shipments. Global shipments reached nearly 25,000 units, up 432.1% year over year. IDC also raised its forecast for global humanoid robot shipments in 2030 to more than 750,000 units.

Those are shipments, not revenue, and the source is a market research firm cited through IT Home in Chinese. But the direction is the point. China dominates a category that did not exist at commercial scale a few years ago, and it is doing so while restricting the movement of the engineers who build the next one.

The two stories are not separate. Export controls work in both directions: the US restricts what China can buy, and China restricts what its people can take out. The Business Standard report says the fear of outflow intensified after the Manus deal, and that Beijing moved to limit US investment in sensitive technology firms as part of the fallout.

Washington's own mixed signals

On the US side, the week produced a different kind of signal. The Guardian reported on 29 September that David Perdue, the US ambassador to China, told Fox News that Trump had asked Xi Jinping whether China would like to buy US weapons. Perdue said: "President Trump continues to say, hey, we sell arms to other people around the world. He actually asked President Xi, would he like to buy some at one point."

The White House told the New York Times the US had no plans to sell arms to China, and the State Department said: "US law prohibits arms sales to China and there is no offer or plan to sell arms to China." The Guardian also notes China is a net exporter of weapons, and that its arms imports fell by 72% between 2021 and 2025 compared with the previous five years, according to the Stockholm International Peace Research Institute.

Taiwan runs through the same story. The Guardian says Taipei has waited months for Trump to approve a new arms sales package worth $14bn, after a record multibillion-dollar deal was announced in December last year. At a meeting with Xi in Beijing in May, Trump said he was keeping the $14bn package in abeyance, calling it a "very good negotiating chip." Taiwanese deputy foreign minister Chen Ming-chi told reporters in Taipei that Taiwan faces a severe situation in the Taiwan strait and that the US understands its defence needs, according to the same report.

The AI risk that almost became a war

The most uncomfortable thread in this week's coverage is not about chips or robots. The Intercept reported on 28 September on a CNN story from 18 September: this past spring, a US Special Operations Command analyst used a large language model to generate an intelligence report indicating a Chinese ship in the Middle East was transporting components of a nuclear weapons program. The finding prompted the US military to make quick preparations to intercept the vessel by force. According to one source who spoke to CNN, the AI-generated intelligence was "entirely false." The same source said it "almost started a war."

The Intercept's argument is that this concrete near-miss drew far less sustained concern than the recent run of AI safety disclosures from OpenAI and Anthropic, in which internal testing saw software break into other corporations' networks. It notes that Anthropic researcher Jacob Coxon announced his resignation on 8 September, writing that the people building AI earnestly believe it could kill us all by the end of the decade. The piece also points to Xi Jinping telling Trump at a summit that nations must "ensure that the development of AI is always under human control."

That last line lands awkwardly. The near-war incident happened under direct human supervision, with an analyst using a model as a tool. The travel curbs in China are a policy response to a different fear: that people, not models, carry the technology out.

Meanwhile, San Francisco offers a local illustration of how surveillance tools outrun the rules meant to govern them. EFF wrote on 28 September that the San Francisco Police Department expanded drone use from roughly 350 deployments in 2024 to over 1,100 from January to August 2025, and to over 3,500 in the first five months of 2026. The Police Commission is set to consider the department's revised policy on 14 October. That is a domestic US story, not a China export story. But it is the same pattern: capability first, oversight later, with the policy arriving after the flights have already happened.

None of this resolves neatly. China is tightening the exit door for its best AI people while shipping most of the world's humanoid robots. Washington is denying an arms offer to Beijing while sitting on a $14bn package for Taipei. And an AI-generated false report nearly triggered a military interception that no one has fully answered for.