China FIREWALL Act of 2026
- Sponsor
- Rep. Gottheimer, Josh [D-NJ-5]
- Committees
- Oversight and Government Reform Committee (primary)
- Last action
- Oct 1, 2026
Bottom line
The China FIREWALL Act aims to secure U.S. government systems by banning AI technology linked to foreign entities of concern, primarily targeting Chinese AI models, while allowing for national security waivers.
What it actually does
This bill prohibits executive agencies from using or procuring artificial intelligence (AI) applications developed or provided by a 'foreign entity of concern,' or any software utilizing a foundation model from such an entity. It mandates the Office of Management and Budget (OMB) to develop standards for removing existing prohibited AI models from government information technology, with exceptions for national security, law enforcement, and security research, provided risk mitigation actions are documented.
Proponents argue
Proponents argue that this bill is a critical step to safeguard national security, protect sensitive government data, and prevent espionage or intellectual property theft by adversarial foreign governments, particularly China. They contend that AI models from foreign entities of concern could contain backdoors, vulnerabilities, or be subject to foreign government influence, posing unacceptable risks to federal operations and intelligence.
Opponents contend
Opponents might argue that such a broad prohibition could stifle innovation within government agencies by limiting access to advanced AI tools, regardless of their specific security posture. They may also raise concerns about the potential for retaliatory measures from foreign governments, the significant cost and logistical challenges of identifying and replacing existing AI systems, and the risk of hindering legitimate research or operational effectiveness if suitable alternatives are not readily available.
The bill is very short and clearly written, making it highly accessible for any reader to understand its core provisions within minutes.
Section 2(a)(1)
Prohibition on Use of Chinese AI Models on Government IT
This provision directs the Director of the Office of Management and Budget (OMB), in consultation with several key agencies (GSA, CISA, DNI, DoD), to establish standards and guidelines within 60 days of the bill's enactment. These standards will require executive agencies to remove any 'covered application' from their information technology systems. This aims to eliminate existing AI models from foreign entities of concern from government use.
Supporters argue
Supporters argue this is crucial for immediately addressing existing security vulnerabilities and data risks posed by foreign-developed AI already integrated into government systems. It provides a clear directive for agencies to secure their digital environments.
Critics contend
Critics might contend that a 60-day timeline for developing comprehensive standards and requiring removal across all executive agencies is overly ambitious and could lead to rushed, ineffective, or disruptive implementation, potentially impacting critical government functions.
Tradeoffs
This provision balances the urgent need for national security with the practical challenges and potential disruptions of rapidly overhauling existing technological infrastructure across numerous federal agencies.
Section 2(a)(2)
National Security and Research Exceptions
This subsection allows for specific exceptions to the prohibition on using covered AI applications. These exceptions are for law enforcement activities, national security interests and activities, and security researchers. For any authorized use under these exceptions, agencies are required to develop and document specific risk mitigation actions, ensuring that even exempted uses are managed securely.
Section 2(b)
Procurement Prohibition
This section explicitly forbids the head of any executive agency from procuring a 'covered application' or any software that incorporates a 'covered application.' This means that after the bill's enactment, federal agencies cannot purchase new AI systems or software that originate from or utilize foundation models developed by foreign entities of concern.
Section 2(c)
Waiver Authority
This section grants the Director of the Office of Management and Budget (OMB), in consultation with the Administrator of General Services, the Director of the Cybersecurity and Infrastructure Security Agency, the Director of National Intelligence, and the Secretary of Defense, the authority to waive the requirements of this section on a case-by-case basis. Such waivers can only be granted 'in the interest of national security,' providing a mechanism for flexibility in exceptional circumstances.
The bill defines 'covered application' as any AI application developed or provided by a 'foreign entity of concern,' or software utilizing a foundation model from such an entity. Crucially, 'foreign entity of concern' is not defined within this bill but by reference to section 40207(a) of the Surface Transportation Reauthorization Act of 2021 (42 U.S.C. 18741(a)). This external definition includes entities owned by, controlled by, or subject to the jurisdiction or direction of a government of a foreign country that is a 'covered nation,' which includes China.
Section 2(d)(2) and 2(d)(4)
Why it matters:This is standard legislative practice to avoid redundancy and ensure consistency with existing legal frameworks, rather than an attempt to obscure. However, it does make the bill less self-contained.
Case for: Supporters would argue that referencing existing definitions ensures consistency across federal law and leverages established criteria for identifying adversarial foreign entities, streamlining the legal framework.
Case against: Critics might argue that relying on external definitions, especially from unrelated statutes, can make the scope of the bill less transparent and potentially subject to changes in those external definitions, leading to uncertainty or unintended breadth.
Estimated impact: This definition is fundamental to the bill's entire scope, determining which AI applications and providers are subject to the ban. Its impact is total, affecting all aspects of the bill's implementation.
The bill defines 'foundation model' as a software component that implements AI, uses computational/statistical/machine-learning techniques to produce outputs, is trained on a broad data set, is designed for generality of output, and is adaptable to a wide range of distinctive tasks. This definition is critical because the prohibition extends not just to 'covered applications' themselves, but also to 'any software that utilizes a foundation model developed or provided by a foreign entity of concern.'
Section 2(d)(5)
Why it matters:This is standard practice for defining key terms within a bill. Its placement is not evasive, but its implications for the bill's reach are substantial and could be overlooked.
Case for: Proponents would argue that defining and including 'foundation models' is essential to prevent circumvention of the ban. Without it, foreign entities could provide core AI technology that is then repackaged by other companies, undermining the bill's intent to secure the AI supply chain.
Case against: Critics might argue that this broad definition could create significant compliance challenges, as identifying the origin of every underlying foundation model in complex software systems can be extremely difficult and costly, potentially leading to unintended prohibitions.
Estimated impact: This definition significantly expands the scope of the ban, potentially impacting a much wider array of software and services than just direct AI applications. It could necessitate deep supply chain audits for all government software.