US Treasury examining Benchmark Capital’s ties to Chinese startup Manus AI

Following criticism of Benchmark by people in the tech industry, the Treasury Department sent the firm an inquiry about the Manus funding, essentially asking the firm if the investment is covered by the new restrictions, or to explain why it is not. Benchmark has not yet responded, according to people familiar with the matter.

It’s likely Benchmark will echo its counsel, one person said, telling the Treasury Department that notification of the investment is not required. Benchmark’s lawyers from multiple large US law firms argue that Manus does not advance the AI field in China because it does not train the models it uses to carry out tasks. Those models are made by Anthropic’s Claude, Alibaba and possibly others, and Manus only fine-tunes some of them.

But Treasury’s enforcement of the new rules has faced few tests, and it is unclear how it will handle this case.

While Manus doesn’t train models, its co-founder and chief scientist Ji Yichao wrote on X last month that the company’s agent has exhibited “emergent capabilities,” suggesting that its technology could be advancing AI.

Last year, Ji, a prominent technology entrepreneur in China who has been on the cover of the Chinese version of Forbes magazine, published a series of open-source AI models that he created, aimed at reproducing OpenAI’s then state-of-the-art o1 reasoning product. “Given the limitations of my abilities, time, and financial resources, it’s unlikely that I’ll be the one to take this effort all the way to the finish line,” he wrote in a blog post at the time. “Therefore, I feel it’s important to share these experiences — worth the equivalent of dozens of H100 GPUs — while they’re still fresh.”

While advances in large language models like OpenAI’s GPT were achieved by training on the most powerful supercomputers in the world, recent improvements in AI capability have come less from training and more through new techniques in inference that are often referred to as “reasoning,” because the models are run multiple times until the optimal response is achieved.

It’s possible that with more funding, Manus could improve AI capabilities by advancing reasoning techniques. It’s unclear whether US regulators would see that as advancing AI, or agree with Benchmark’s analysis that it is simply employing AI that has already been built.

Benchmark is one of Silicon Valley’s most successful venture firms, known for making carefully selected bets on relatively few startups. It provided early capital to companies like eBay, Twitter, Uber and Snap.

Because of Benchmark’s high profile and reputation, its Manus investment raised eyebrows in tech’s more hawkish circles, especially as others were going in the opposite direction. Sequoia Capital, another storied venture firm, cut ties with its successful Chinese arm after years of criticism that helped spur the Biden Administration’s outbound investment rules.

There is widespread worry that Chinese technology could provide a back door to Beijing. And in the heart of Silicon Valley, there are new allegations that Chinese nationals studying at Stanford, particularly in the field of AI, are being pressured to act as spies for China.

Even startups, which have widely adopted DeepSeek’s open-source AI models, are being criticized for using technology that may have unknown security vulnerabilities that could potentially benefit China.

The Trump administration is considering even tighter restrictions on outbound investments to China to ensure more dollars don’t slip through the regulatory cracks.

Benchmark general partner Bill Gurley, one of the industry’s best-known venture capitalists, has criticized moves to cut ties with China.

“Our nation’s recent curbs on Nvidia H20s, intended to slow China AI innovation, will enhance & accelerate Chinese AI innovation. This action will create the OPPOSITE of the intent,” he posted last month.

Gurley did not respond to a request for comment.

Critics of US restrictions on exports to China make some good points. Starved of high-end AI chips and US investment dollars, Chinese AI researchers have nonetheless innovated in the field.

In some ways, a lack of resources has forced researchers there to think creatively, learning new techniques to train and run models more efficiently.

But even those advances in efficiency have relied on distilling models built by US tech companies.

It has advanced its homegrown chip industry, but it was doing that anyway, as part of a massive effort to end reliance on US technology, in part by allegedly engaging in a sprawling espionage effort to steal American trade secrets.

The export restrictions have likely slowed China’s advances in AI — how much is unclear. It’s possible they have only bought the US a slightly larger lead in the technology race.

There is really only one way the US can win the race in the long term, and that is by relentlessly innovating at a pace that China cannot hope to match.

This was the playbook in the Cold War against the Soviet Union. It’s a strategy that involved a boatload of government research dollars. Inexplicably, the Trump White House has disrupted crucial funding in those areas, placing a giant hurdle in the way of US innovation, just as the race is getting closer.

A winning US strategy involves both slowing China as much as possible and doing everything it can to accelerate innovation at home.

Benchmark may have a legitimate argument for why its Manus investment is in the US interest — $75 million isn’t enough to do much in AI research these days — but it should still be vetted for possible national security implications. The fact that the firm chose not to initially notify the Treasury Department of its investment in Manus suggests there’s something wrong with the process.

In the past, US government reviews on inbound investment have taken incredibly long, and it’s a legitimate worry that an analysis of an outbound investment could be equally vexing. There should be some way that a vetting process can coexist with the fast-paced tech industry, where even a delay of a single day could derail an investment.

The same is true with visas for students and tech workers coming to the US. During the Cold War, the US was the recipient of incoming scientific talent from hostile nations, and authorities figured out a way to mitigate the risks of an open system. There’s no reason that can’t happen this time around.

When asked whether Benchmark’s investment into Manus is in the US national interest at a Senate hearing Thursday, Microsoft President Brad Smith said “the number one comparative advantage of the United States throughout the 50 years that have defined digital technology has been bringing the world’s best people to our country and giving them access to venture capital.

“And we should continue to burnish both of those. And I think you’re right to ask where else is venture capital going. I’ll just say this. If we can keep bringing the best people to the United States and if we can keep educating the best people in the United States, I believe the money will be here to help them succeed. But let’s make sure we’re continuing to bring the best people in the world and giving them the opportunity to build great companies here in the United States.

“There’s a really good question about whether [US investments in companies like Manus helps the country]. I recognize that you all are quite rightly focused on that. I’ll just keep saying: Bring the people here and they will have access to the money and we’ll continue to outcompete the world.”