The U.S. Treasury Department is reviewing a $75 million investment made by Benchmark Capital into Manus AI, a startup with operations linked to China. This marks another flashpoint in the escalating tech rivalry between Washington and Beijing—especially over the future of artificial intelligence.
Benchmark, a top-tier Silicon Valley venture firm, was recently contacted by the Treasury as part of a review under new rules aimed at curbing U.S. investment in sensitive technologies in “countries of concern.” These restrictions fall under the Outbound Investment Security Program, launched after a 2023 executive order by President Biden. Though the law only took effect earlier this year, it requires U.S. investors to notify the Treasury about investments in sectors like AI that could strengthen foreign military or surveillance capabilities.
Is Manus AI Truly a Threat?
Benchmark reportedly received legal guidance from multiple U.S. law firms stating that its investment in Manus AI doesn’t fall under the new rules. According to these legal teams, Manus doesn’t develop its own foundational AI models. Instead, it’s considered a “wrapper”—a company that builds tools around existing models from Anthropic, Alibaba, and others. Manus also fine-tunes models but does not train them from scratch.
Additionally, the startup isn’t officially Chinese, its lawyers argue. Its parent company, Butterfly Effect, is incorporated in the Cayman Islands. While some of its workforce operates in China, it also has employees in Singapore and Japan. Crucially, all company data is stored on servers outside China, managed by Western cloud providers.
Still, some prominent voices in the venture ecosystem aren’t convinced. Founders Fund partner Delian Asparouhov criticized the move, saying it “makes zero sense,” and Lux Capital’s Josh Wolfe echoed similar concerns on social media. The central fear: even seemingly indirect U.S. investments could give China an edge in AI development.
Rising AI Capabilities Raise Red Flags
Manus raised eyebrows earlier this year with a viral demo that showcased its AI agent autonomously handling complex tasks—from building mobile apps to conducting deep research. This was dubbed a “second DeepSeek moment,” referencing another Chinese-linked AI breakthrough that rattled global markets.
Though Manus doesn’t train its own large language models, its co-founder and chief scientist Ji Yichao has a storied history in AI development. Last year, Ji open-sourced a series of models mimicking OpenAI’s reasoning abilities, noting that resource limitations prevented him from pushing the work further. His recent posts claim the Manus agent is now exhibiting “emergent capabilities,” hinting at potentially significant advances in AI reasoning.
Such developments muddy the waters. Treasury officials must decide whether Manus’s work—despite not involving foundational model training—still poses a risk by accelerating AI capability via inference and reasoning techniques.
A Broader Battle Over Tech and National Security
The Manus review underscores the larger shift happening in Silicon Valley. Top firms like Sequoia Capital have already split from their China arms due to geopolitical pressure. Meanwhile, accusations continue to swirl around Stanford-based Chinese nationals allegedly being recruited as spies. There’s also growing concern that open-source Chinese models like DeepSeek may contain hidden vulnerabilities that could benefit the Chinese government.
Amid this backdrop, some believe the U.S. is falling behind. China is publishing more AI research and launching headline-grabbing models while U.S. regulators work to patch loopholes. Critics argue that without a matching acceleration in domestic innovation, restrictions alone won’t be enough.
Benchmark’s own Bill Gurley criticized chip export controls on platforms like Nvidia’s H20, arguing they may accelerate—not slow—China’s innovation. Others say restrictions are still necessary, but they must be paired with aggressive funding of U.S.-based research, talent, and infrastructure.
The Stakes for Innovation and Security
Manus AI’s case may serve as a test of how the U.S. applies its outbound investment rules. Treasury has had limited experience enforcing the new law, and this case could help define its future approach. Will investments in “wrappers” or inference-based platforms be treated with the same scrutiny as foundational model labs?
Microsoft’s President Brad Smith weighed in at a recent Senate hearing, emphasizing that America’s greatest edge has always been its ability to attract top global talent and pair it with venture capital. Rather than shut doors, he argued, the U.S. should focus on welcoming innovators while applying smart oversight.
But that balance—between open innovation and national security—is proving hard to strike.
