On June 1, NVIDIA unveiled its first open humanoid reference design for academic robotics, built on Unitree’s H2 Plus chassis, with named users including Stanford, ETH Zurich, UC San Diego, and Ai2. Two days later, a House bill moved to put that class of robot on a list that would block it from the US market.
The bill is the Guarding the US Against Adversarial Robotics Dominance (GUARD) Act, introduced on June 3 by Rep John Moolenaar, chair of the House Select Committee on the CCP, with Reps. Jay Obernolte and Jennifer McClellan. This post is a detailed read of what this bill means.
What it does
GUARD directs national security agencies to evaluate humanoid and quadruped robots, and the software and communications systems that power them, when they are produced by entities tied to a country of concern.
If an agency finds the robot poses an unacceptable risk, the FCC places it on the Covered List within 30 days, i.e. gets banned. If no agency makes any determination within a year of enactment, every covered robot is added automatically.
Inaction defaults to the ban.
What counts as a covered robot
The definition is functional and wide:
The term ‘‘humanoid or quadruped robot’’ means—(A) a mechanical device that—(i) possesses a body structure that uses 2 or 4 articulated limbs for locomotion, navigation, or movement on the ground; (ii) operates at a distance from a human operator or supervisor autonomously, semi-autonomously, based on commands or response to sensor data or any combination thereof; and (B) any external device designed to control a mechanical device described in subparagraph (A)
Two or four limbs, some autonomy, and the controller counts too. That captures most of what ships in the category today.
Who counts as a covered entity
GUARD does not write its own adversary list. It points to “country of concern” as defined in 22 U.S.C. 2651a(m)(1), which names six: China, Russia, Iran, North Korea, Cuba, and Syria. The operative term is country of concern, not foreign adversary, and the two lists are not identical.
The entity test is where the line actually sits (pages 5-6):
A) an entity that—(i) is headquartered in, has its principal place of business in, or is organized under the laws of a country of concern; or (ii) is subject to the influence, direction, or control of the government of a country of concern;
(B) an affiliate or wholly or partially owned subsidiary of an entity described in sub-paragraph (A);
(C) an entity in a joint venture with an entity described in subparagraph (A); and (D) an entity with which an entity described in subparagraph (A) has a technology sharing or licensing agreement.
The test is about who produces the robot, not where it is assembled. Final-assembly location does not launder a covered entity. It also reaches outward: prong (C) covers a joint venture partner, and pront (D) covers technology sharing or licensing agreement.
Entity and equipment, and where it gets murky
The framework is entity-based and equipment-based at once.
If the producer is Unitree, its robots are blocked for future sale no matter where the final assembly happens.
If a US firm assembles a robot using Chinese control software, radios, or modules, the communications equipment and where the components originated can still make them eligible for listing.
The drone precedent
Two reference points from the drone ban can be useful as a reference to this GUARD bill.
(1) under Federal Acquisition Regulation (FAR) origin tests (the rulebook for US government purchasing, which judges where a product counts as made by component cost share), foreign components can pull a whole system into scope.
(2) the Defense Innovation Unit’s Blue UAS list offers an escape valve, a vetted whitelist of drones certified to have no covered-nation supply chain and cleared for government use. A domestically assembled robot with vetted, non-covered communications could seek an equivalent path.
Where exactly that line falls is left to agency rulemaking rather than written into the statute, so the real scope of the ban stays unknown until well after passage.
Process and likelihood
Since it was introduced June 3 in the House of the 119th Congress and referred to committee, GUARD would need a committee markup, a House floor vote, passage through the Senate, reconciliation of any differences between the chambers, and a presidential signature.
Standalone near-term passage is unlikely. Realistically this attaches to the FY2027 NDAA rather than passing alone.
Three things raise the odds above a typical bill: (1) There is bipartisan China-hawk consensus behind it. (2) The DJI case already proved the Covered List mechanism works for this purpose. (3) the FCC may not wait for Congress. There’s a May 2025 committee letter already urging Unitree’s Covered List and Entity List designation and the agency has the standing authority to act.
What a ban would cost US robotics
Unitree robots are already embedded across the US research base. Per its own pre-IPO filing, the company shipped more than 5,500 humanoid and 23,000 quadruped units in 2025. Its robots sit in university labs, private companies, and public safety fleets.
There is no domestic substitute at the price point. The G1 sells at $16K and R1 $6K. The nearest non-Chinese humanoid a lab can actually buy runs well into six figures. And for quadrupeds, a Boston Dynamics Spot costs 20-45X of a Unitree Go2.
In the near term, a ban does not redirect that demand to US hardware, since the US hardware does not exist at that price. It pushes labs toward smaller fleets, simulation, or open-source builds.
It also raises the question of whether some hardware-dependent research migrates to collaborators offshore and routes back. A national-security bill that ends by moving US robotics work onto foreign soil would be an odd result, and it is a real one to watch.