Washington Turns AI Capital Into a Weapon

The rule is not a ban on China. It is a ban on certain forms of American help.

Washington’s newest outbound-investment proposal looks technical, but its political meaning is blunt: the United States is no longer treating capital as neutral when it flows into strategic technology. In the old globalization story, money crossed borders in search of return. In the new security story, money carries know-how, credibility, networks, board access and managerial discipline. The U.S. Treasury now wants to police those intangible benefits when they reach companies tied to China, Russia, Iran, North Korea, Cuba and Venezuela’s Maduro-linked regime.

On October 22, 2026, Treasury issued a proposed rule to implement the Comprehensive Outbound Investment National Security Act, known as the COINS Act; the Federal Register version was published on October 26 and comments are due on December 7, 2026. The proposal would preserve the basic architecture of the U.S. Outbound Investment Security Program — some deals must be notified to Treasury, some are prohibited — but would expand both the geography and the technologies covered. The most politically charged change is that artificial intelligence appears under the Trump administration’s preferred label, “super intelligence,” and the proposal would treat development of any covered SI system as prohibited rather than merely notifiable. (home.treasury.gov)

This is not yet final law. The current Part 850 outbound-investment rules remain in force until final regulations set a new effective date. But serious investors do not wait for the last page of the Federal Register. Venture funds, corporate investment arms, chip companies, pharmaceutical groups and non-U.S. funds with American limited partners are already being forced to ask a basic question: can a dollar invested in a Chinese technology company become evidence in a national-security file?

What would change?

The proposal expands the outbound-investment regime in four ways.

First, it broadens the country list. The Biden-era executive order and the current rules focused on China, including Hong Kong and Macau. The COINS Act framework adds Cuba, Iran, North Korea, Russia and Venezuela under Nicolás Maduro’s regime. Treasury’s proposed rule follows that statutory list. (thefederalregister.org)

Second, it expands the technology list. The COINS Act authorizes rules for semiconductors and microelectronics, AI systems, quantum information technologies, high-performance computing and supercomputing, and hypersonic systems. Treasury also proposes to add biotechnology, using its authority to define additional technologies that may enable military, intelligence, surveillance or cyber-enabled capabilities of a country of concern. (home.treasury.gov)

Third, it hardens the treatment of AI. Under the existing outbound-investment regime, some AI-related investments are prohibited, while others are notifiable depending on thresholds and end uses. Treasury’s proposal says that rapid advances in “super intelligence” capabilities justify a broad prohibition on U.S.-person investment in development of SI systems in covered foreign persons. The proposed definition is much broader than science-fiction “superintelligence”: it covers machine-based systems that generate content or make predictions, recommendations or decisions, and also software, hardware, tools, agent harnesses, robots, autonomous vehicles and unmanned systems that operate using such systems. (thefederalregister.org)

Fourth, the rule pushes due diligence onto investors. Treasury describes the system as a self-assessment regime: a U.S. person must decide whether a transaction is prohibited, notifiable, excepted or outside the rule. Notifications would generally be due within 30 calendar days after completion, and also within 30 days after later learning facts that would have made the deal notifiable or prohibited. (thefederalregister.org)

The $50,000 threshold is not the loophole many readers may think it is

The Associated Press reported that the proposal would apply to investments exceeding $50,000 and would exclude many financial investments, including stock purchases. That summary is directionally useful, but the legal detail is more important. Treasury’s text says the $50,000 de minimis exception applies to transactions that would otherwise be notifiable — not to prohibited transactions. Treasury explicitly says prohibited transactions below that value can still raise national-security concerns. (apnews.com)

That distinction matters. A wealthy individual writing a modest check into a covered AI developer cannot assume that the transaction is harmless because it is small. A passive purchase of a publicly traded security is treated differently from venture capital because venture capital is often not just money. It can include advice, recruitment help, strategic introductions, technical validation and access to customers. Treasury and outside analysts repeatedly focus on those non-cash benefits.

There are important carve-outs. The proposal excepts publicly traded securities, securities issued by SEC-registered investment companies such as mutual funds or ETFs, derivatives of specified securities, certain limited-partner fund investments, bank lending, payment processing, underwriting, prime brokerage, global custody, equity research and ordinary or administrative business transactions. Limited partner commitments of up to $2 million may qualify for an exception, and a larger LP may rely on binding contractual assurances that its capital will not be used for covered prohibited or notifiable transactions. But governance rights beyond standard minority protections can destroy the exception. (thefederalregister.org)

In plain English: Treasury is not trying to ban every American from owning a China-linked ETF. It is trying to restrict American participation that looks like strategic financing, governance, scaling support or deal engineering in sensitive sectors. That is a narrower claim than “financial iron curtain,” but it is also more intrusive than a simple sanctions list.

Why Washington is doing this: the evidence is real, but incomplete

The strongest argument for the rule is not ideological. It is structural. Advanced AI, chips, quantum computing, supercomputing, hypersonics and biotechnology are dual-use technologies. They can produce consumer products, medical breakthroughs and productivity gains; they can also support military targeting, cyber operations, surveillance systems, weapons development and state coercion.

Treasury’s preamble says that since the outbound program began in January 2025, it has received about 100 notifications of investments in semiconductor and AI industries from venture capital firms, asset managers, investment banks, individual investors and entrepreneurs. It also says commercially available data showed almost 230 unique U.S.-based investors completed about 300 initial and add-on transactions over three years in the covered sectors in countries of concern. These are not enormous numbers compared with the whole U.S. capital market, but they are large enough to matter in frontier sectors where a single fundraise can decide whether a company survives. (thefederalregister.org)

Independent research gives Washington more ammunition. Georgetown’s Center for Security and Emerging Technology found that from 2015 to 2021, 167 U.S. investors participated in 401 investment transactions into Chinese AI companies, representing 17% of 2,299 global investment transactions in Chinese AI firms. Observed transactions involving U.S. investors totaled $40.2 billion into 251 Chinese AI companies, though CSET cautioned that it could not identify exactly how much of that amount came from U.S. investors. CSET also found that 91% of observed U.S.-linked transactions were at venture-capital stages. (cset.georgetown.edu)

That caveat is crucial. The public evidence does not prove that each U.S.-linked investment strengthened China’s military or surveillance apparatus. It proves that American investors were present in China’s AI financing ecosystem and that the precise U.S. share is hard to measure. In a national-security bureaucracy, uncertainty often becomes an argument for more control. In a free-market system, uncertainty should also become an argument for clearer evidence and narrower rules.

The money: this is small compared with global AI capital, but large where it counts

The United States dominates outgoing AI venture capital. The OECD’s 2026 report on AI venture investment estimated that U.S. VC investors represented about 56% — USD 124 billion — of worldwide outgoing VC investment in AI firms in 2025 where investor country could be identified. The United Kingdom followed with 9% or USD 20.7 billion, China with 8% or USD 17.2 billion, and EU27 investors with 7% or USD 14.5 billion. (oecd.org)

That concentration is the quiet foundation of the rule. If American capital were marginal, Washington’s restrictions would be symbolic. Because U.S. investors remain central to AI finance, the rule can shape market behavior even outside U.S. territory. Non-U.S. funds with American LPs, foreign managers employing U.S. decision-makers, and Chinese companies seeking Western legitimacy all have to price in the possibility that American participation is toxic or administratively burdensome.

Rhodium Group’s earlier analysis shows why the debate is not limited to new AI startups. U.S. Bureau of Economic Analysis estimates cited by Rhodium put total assets of U.S. multinational companies in China at $708 billion at the end of 2020. Rhodium also found that U.S. technology financing in China peaked at $14.4 billion in 2018 and dropped to a 10-year low of $1.27 billion in 2022, with investors increasingly avoiding national-security-sensitive deep tech while much money moved toward biotechnology and consumer-focused companies. (rhg.com)

This helps explain why biotechnology suddenly matters. If investors have already retreated from chips and quantum because those sectors are politically radioactive, capital will search for adjacent areas. Biotech can be civilian, lucrative and lifesaving; it can also touch genomics, bio-manufacturing, brain-computer interfaces, dual-use lab automation and supply-chain dependency. Treasury’s proposed addition of biotechnology is therefore not random. But its breadth is what should worry anyone who wants national security rules to be precise.

The AI language problem: “super intelligence” as policy branding

The proposed rule’s terminology is not a minor stylistic issue. The Federal Register footnote says Executive Order 14434, issued on September 29, 2026, requires executive departments and agencies to use “Super Intelligence” and “SI” in place of “Artificial Intelligence” and “AI” in certain documents. (thefederalregister.org)

This is politically useful and analytically dangerous. “Super intelligence” sounds existential, futuristic and alarming. But Treasury’s proposed definition covers ordinary machine-learning systems that generate content or make predictions, recommendations or decisions. It also reaches tools and applications operating in whole or in part using such systems, including industrial robots, humanoid robots, autonomous vehicles and unmanned systems. (thefederalregister.org)

The risk is not that words alone create repression. The risk is that inflated words make broad prohibitions easier to sell. If a policy is about “AI applications,” readers may ask for evidence. If it is about “super intelligence,” the political imagination jumps to machines beyond human control. A serious national-security regime should not rely on branding. It should define the technical risk, the transaction risk and the evidence threshold with surgical clarity.

WilmerHale and Morgan Lewis, both writing from the perspective of legal compliance rather than civil liberties activism, reached similar practical conclusions: the AI/SI change is a major expansion. WilmerHale says the proposal would prohibit investments in Chinese companies that develop any AI/SI system, replacing the current tiered approach. Morgan Lewis says covered transactions involving SI development would be prohibited regardless of training-compute thresholds and end uses, and that investors should examine application developers, not just model developers. (wilmerhale.com)

The China argument — and what Beijing omits

China’s public response is predictable but not irrelevant. Asked about the biotechnology proposal, Foreign Ministry spokesperson Mao Ning said measures targeting specific countries and restricting economic, trade and investment activity by “overstretching” national security serve no one’s interest; she urged the U.S. to respect market-economy principles and normal business cooperation. AP reported a similar response and added that China said it would closely follow developments and safeguard its legitimate rights and interests. (mfa.gov.cn)

Beijing has a point when it warns that national security can become elastic. Every great power is now tempted to describe industrial policy as security policy. But China’s argument is incomplete. It does not answer the specific U.S. concern that venture capital can transmit intangible support to companies operating in surveillance, military modernization or cyber-relevant sectors. Nor does it confront the fact that China itself uses industrial planning, data controls, export controls and state direction to shape strategic technology markets.

The proper conclusion is not that Washington is innocent and Beijing is guilty, or the reverse. The proper conclusion is that both powers are moving away from open technology globalization and toward managed technological blocs. The U.S. calls it national security. China calls its response protection of legitimate rights. Investors call it compliance risk. Ordinary people will experience it as more expensive technology, fewer open research channels and a world where scientific collaboration increasingly depends on passports and political trust.

Europe: slower, softer, but moving in the same direction

The European Union has not copied Washington’s outbound-investment ban. It is moving more cautiously. The Commission says its January 2025 recommendation asks member states to review outbound investments into non-EU countries in semiconductors, AI and quantum technologies, including ongoing and past transactions going back to January 1, 2021. Member states were asked to report implementation and risks by June 30, 2026. The EU’s stated objective is to determine whether further action is needed to prevent outbound investments from harming economic security. (policy.trade.ec.europa.eu)

This is an important difference. The U.S. is proposing prohibitions. The EU is still largely in assessment mode. But the direction is similar: advanced technology investment is no longer treated as ordinary commerce. Europe’s dilemma is sharper because it wants strategic autonomy without becoming merely an annex of U.S. tech policy or a passive market for Chinese technology. Its problem is that it lacks U.S.-scale AI capital and compute power. A world of investment walls may protect Europe in theory while weakening it in practice if its firms cannot scale.

What could go wrong?

The case for targeted outbound controls is serious. It would be irresponsible to pretend that AI chips, autonomous systems, bioengineering platforms and hypersonic technologies are ordinary consumer goods. But the proposal carries at least five risks.

First, overbreadth. If “super intelligence” captures too many ordinary AI applications, Treasury may end up chilling benign investment without clearly reducing military risk. A robotics startup, logistics AI company or medical AI tool can be civilian, military-relevant, or both. The rule’s practical effect will depend on definitions and enforcement judgment.

Second, concentration. Restrictions that large funds can manage may crush smaller investors and startups under compliance costs. The largest technology companies, law firms and defense-linked incumbents can afford screening systems. Smaller challengers may avoid entire markets. A rule designed to limit China could also consolidate AI power inside the United States among the few firms able to navigate the bureaucracy.

Third, retaliation and fragmentation. China can respond with its own controls on technology exports, licensing, data, critical minerals or pharmaceutical collaboration. Morgan Lewis notes that China has separately proposed additions to its technology export catalogue covering AI, biopharmaceutical and photovoltaic technologies, with an interim registration requirement already operative for some exports. (morganlewis.com)

Fourth, evidence opacity. Treasury may receive confidential information and may eventually operate databases or guidance mechanisms, but ordinary citizens will not see most of the evidence behind many decisions. That may be unavoidable in national security. It is also dangerous because secrecy protects both legitimate intelligence and bureaucratic overreach.

Fifth, humanitarian spillovers. Biotechnology is the clearest example. Treasury itself acknowledges, according to Morgan Lewis’s reading of the proposal, that biotech restrictions could have countervailing public-health or commercial effects. A badly drawn rule could slow drug collaboration, reduce investment in useful therapies or push research into less transparent jurisdictions without stopping genuinely dangerous work. (morganlewis.com)

What this means for ordinary people

This debate sounds remote: venture funds, Federal Register pages, export-control lawyers. It is not remote.

If Washington is right, the rule may reduce the chance that American capital helps build systems used for military targeting, mass surveillance, cyberattacks or biological threats. That matters to soldiers, dissidents, journalists, minorities, patients and citizens whose lives can be shaped by technologies they never chose.

If Washington is wrong or overbroad, the rule may accelerate a divided world in which medical research, AI tools, chips and data systems split into rival blocs. Consumers may pay more. Scientists may collaborate less. Startups may find fewer investors. Governments may gain another lever to decide which technologies may cross borders and which people may participate in building them.

The deeper issue is human sovereignty. AI power is already concentrated in a small number of states and corporations. Outbound-investment controls may be justified when they block concrete military or surveillance risks. But if they become a permanent architecture of technological nationalism, they may also strengthen the very concentration of power they claim to discipline.

The unanswered questions

Before Treasury finalizes the rule, the public should demand answers to several questions.

What evidence distinguishes a prohibited AI/SI system from an ordinary commercial AI application? Why is a broad prohibition better than a threshold-based or end-use-based system? How will Treasury avoid punishing passive capital while missing non-capital forms of cooperation such as licensing, cloud access, consulting or talent flows? Will biotechnology be narrowed to genuinely high-risk categories, or will ordinary drug development be caught in the net? How will the U.S. coordinate with allies that do not share identical rules? And will Congress receive enough public reporting to judge whether the program actually reduces national-security risk?

The strongest version of the rule would be narrow, evidence-based, technologically precise, internationally coordinated and transparent about aggregate outcomes. The weakest version would become a political instrument: broad enough to scare investors, vague enough to expand, secret enough to avoid scrutiny, and useful enough to consolidate power at home.

Washington says it is preventing adversaries from exploiting American capital. That may be true in some cases. But capital controls are never just about adversaries. They also decide who may build the future, who may finance it, and which technologies become too strategic for ordinary democratic debate.