The American AI Security Act is a bill announced on September 18, 2026 by Representatives Josh Gottheimer (D-N.J.) and Mike Lawler (R-N.Y.) that would establish a mandatory, pre-deployment national-security review of the most powerful AI models, conducted by the National Security Agency. Gottheimer announced it at Fairleigh Dickinson University in Teaneck, New Jersey, with the China FIREWALL Act and a letter to National Cyber Director Sean Cairncross (Source: gottheimer.house.gov).
| Field | Value |
|---|---|
| Chamber | U.S. House of Representatives, 119th Congress |
| Sponsors | Rep. Josh Gottheimer (NJ-5); Rep. Mike Lawler (NY-17) |
| Announced | September 18, 2026 |
| Bill number | Not stated in the announcement |
| Reviewing agency | National Security Agency |
| Status | Announced; bill text not held |
Provisions as described by the sponsors
The account below rests on the sponsors' announcement; the bill text has not been published or captured, and the announcement does not define a "covered model."
- Pre-deployment access. Before releasing a covered model, a developer would be required to give the National Security Agency full access to evaluate whether the model could be used to launch a serious cyberattack or help build a chemical, biological or radiological weapon. Gottheimer's remarks also name attacks on critical infrastructure.
- Time limit. The review would run for 30 days, with a single 30-day extension if needed.
- Developer participation. Companies could provide technical support for the reviews; Gottheimer described the review as one "that other frontier developers can help coordinate," with "industry input."
- Appeal. Developers could appeal decisions through an expedited process modeled on the courts already used for national-security cases.
The announcement does not state what follows an adverse finding, whether release would be barred pending review, or what penalty applies to a developer that does not submit a model (Source: gottheimer.house.gov).
Stated rationale
Gottheimer said "there are no mandatory requirements for pre-deployment testing of frontier AI models," and that the country is "solely reliant on the Trump Administration's voluntary cybersecurity review process based on a June 2026 Executive Order" (the order signed June 2, 2026, EO — Promoting Advanced AI Innovation and Security (Trump, signed June 2, 2026)), which he said had "caused mass confusion in the AI world – nobody really knows who's covered, what the rules actually are, or what happens if you don't play along." He rejected reliance on developers' own testing or on third-party testers and auditors, citing their small number, limited people, compute and access, lack of security clearances and partial funding by the laboratories: "it's like having the fox guard the hen house." He also rejected stopping AI development: "The answer isn't, like some have suggested, to just shut it all down. That's not tenable." Lawler said the bill "takes a commonsense approach: before the most powerful AI models are deployed, we should know whether they could be used to launch devastating cyberattacks or facilitate the development of chemical, biological, or radiological weapons" (Source: gottheimer.house.gov).
Relation to other proposals
The bill places pre-deployment testing with an intelligence agency. Other 2026 proposals place it with the Commerce Department's Center for AI Standards and Innovation, with national laboratories, or with private evaluators accredited by government or engaged by developers; the September 2026 debate over which institution should evaluate frontier models is set out at AI Pre-Release Vetting. Its rejection of third-party testers contrasts with the independent verification organizations of California's SB 813 and the FRONTIER Act (Independent Verification Organizations (IVOs)) and with Anthropic's embedded evaluators (Embedded Evaluation).
Earlier in September 2026 Gottheimer introduced the Stop Rogue AI Act, which directs NIST to set AI-agent security standards and which coverage describes as a Gottheimer–Lawler bill, although Lawler's name does not appear on the copy of the text held (Stop Rogue AI Act — bill text as introduced (119th Congress)).
Open questions
- The bill number, the text, and the definition of a covered model.
- Whether the review would block release, and what enforcement applies.
- How the NSA review would relate to the Commerce Department's existing voluntary arrangements and to any Senate framework.
Relationships
- instance-of: AI Pre-Release Vetting — mandatory pre-deployment government review
- related: National Security Agency (NSA) — the reviewing agency
- related: Josh Gottheimer — lead sponsor
- contradicts: Embedded Evaluation — the sponsors reject reliance on third-party testers funded in part by the laboratories
- related: Stop Rogue AI Act (2026) — earlier Gottheimer bill on AI-agent security standards