Treasury's New Investment Rule Would Treat Nearly All AI Development as a China Risk
The U.S. Treasury Department proposed a rule on October 22 that would rewrite how the government screens American investment in Chinese technology, and the biggest change isn't about new money, it's about who counts as a target. Under the current Outbound Investment Security Program, AI investments are sorted by training-compute thresholds and intended end-use into unregulated, notifiable, or prohibited buckets. The new rule would scrap that tiered system entirely for what it calls "super intelligence systems," replacing it with a binary test: a transaction is either prohibited or it falls outside the rule altogether. There is no longer a middle, notifiable category for AI.
Why Treasury says the old thresholds don't work
Treasury's rationale is that a covered super intelligence system gets captured "regardless of performance or intended use," and that development of one, along with anything that "enhances or extends" its capability, is presumptively off-limits when a covered foreign person is involved — a broader standard than trying to keep recalibrating a compute cutoff as models improve. Treasury Secretary Scott Bessent framed the move as a national-security priority, saying the administration is "committed to putting America's national security first." Assistant Secretary Christopher Pilkerton, in Treasury's own announcement, said the rule "will enhance the Outbound Investment Security Program and our efforts to protect U.S. national security."
A wider net than the sector name suggests
The COINS Act, folded into the fiscal year 2026 defense authorization law, named five covered sectors: semiconductors, AI, quantum computing, high-performance computing, and hypersonics. Treasury's proposal adds a sixth, biotechnology, and separately adds Cuba, Iran, North Korea, and Russia to China as covered countries of concern (Venezuela was reportedly considered and left out). But it's the AI provisions that attorneys have flagged as the most consequential, because the rule's reach extends well past model developers. A client alert from Akin Gump Strauss Hauer & Feld, published this week, says the super-intelligence-system definition covers agentic software harnesses, industrial and humanoid robots, autonomous vehicles, and unmanned systems whenever they plausibly enhance or extend an underlying model's capability. Building an ordinary application on an unmodified third-party model used to sit outside the old rules; under the proposal, that application layer alone could be a prohibited transaction.
Lawyers warn of collateral damage to ordinary companies
Akin Gump's alert specifically works through how the rule could sweep in healthcare and biotech, manufacturing, energy, and financial-services companies that build AI into their products, if those companies carry the wrong ownership links to China — sectors far removed from frontier AI labs but increasingly built on AI tooling. The rule carves out narrow exceptions, including limited-partner commitments of $2 million or less in aggregate and a $50,000 de minimis threshold for notifiable-only transactions (a separate Treasury summary describes the de minimis calculation as an aggregated 12-month figure across covered transactions, a detail worth confirming once the rule text is final), plus ordinary trading in publicly listed securities and funds. Prohibited AI transactions get no de minimis carve-out at all.
What happens next
The rule is not yet in effect and would not apply retroactively; the existing Outbound Investment Security Program regulations stay in force until Treasury finalizes new ones. Written comments are due December 7, 2026, and the COINS Act gives Treasury until March 2027 to issue final regulations. Law firms tracking the docket expect the agency to revise the AI provisions in response to comments, noting that Treasury softened earlier versions of the outbound investment program after a previous comment period. Penalties for violations are steep: civil fines of up to twice the value of the investment, plus potential forced divestment. For venture firms, AI startups with any China-linked ownership, and multinational tech companies doing diligence on acquisition targets, the five-week comment window is likely to be the last real chance to narrow a rule that, as proposed, treats building AI products as itself a national-security question.