"Negotiating AI Safety Globally" is a two-page outline of proposed legislation published by the office of Representative Sam Liccardo (D-California). The document itself is undated and names no sponsor; the office's press release of September 22, 2026 links it as "the full framework" of a bipartisan bill that Liccardo and Representative Kevin Kiley (I-California) would introduce that day (Source: liccardo.house.gov). The outline has two parts: a one-page rationale and a five-section "Outline of Proposed Legislation." It would direct the Secretary of State to seek negotiations with China toward a binding frontier-AI safety agreement, establish a U.S.–China hotline on AI risk, fund an international standards initiative at the Center for AI Standards and Innovation (CAISI), let U.S. experts talk to Chinese counterparts under a conditional safe harbor, and direct separate Commerce negotiations on model distillation. It is an outline, not bill text; the legislation is described at Liccardo–Kiley U.S.–China frontier AI safety negotiations bill (2026).
Rationale
The first page opens by citing an open letter signed by "more than 1300 employees of US frontier AI companies—including many of their most prominent scientists," which warned that "automated AI research could soon accelerate their models' capabilities faster than humans' ability to control them" and called for an international effort to develop the technical means to address that risk and "pace" the frontier. The outline does not name the letter; the description matches the Pacing the Frontier statement of July 28, 2026, reported as having more than 1,300 signatories by August 9. The outline answers: "They're right."
It argues that domestic safety bills "remain insufficient when developers in the U.S. and the rest of the world increasingly rely upon open-source Chinese models beyond the reach of American law," and frames the obstacle to domestic regulation as a question it says Congress cannot answer: "How can we make Americans any safer with U.S. regulation that ties the hands of American AI labs, but cannot reach hundreds of millions of Chinese open-source models that produce nearly as powerful (and at least as dangerous) AI agents?" To halt "a 'race to the bottom' on AI safety," it calls for a shared framework under which developers in both countries adhere to the same standards and constraints, "paired with strong verification measures," and predicts that "whatever safety rules and standards the U.S. and China can settle upon, the rest of the world will likely follow—or suffer technological exclusion."
The outline states that the proposal "carries no delusions about this Administration's current willingness to lean in to safety regulation, or to any negotiation with Beijing on the topic." Its stated premise is that technical groundwork can proceed regardless: experts should "find common ground around metrics, standards, protocols, and the like," and industry, civil society and academia "should begin working with their foreign counterparts now—starting with those in China," so that standards are ready "whenever that policy window opens."
Outline of proposed legislation
Section 1 — State Department negotiations. The Secretary of State, in consultation with the Secretaries of Commerce, Treasury and Defense, the Director of National Intelligence and others the Secretary deems appropriate, "shall seek to commence negotiations with the government of China" to:
- identify shared priorities for mitigating frontier AI safety and catastrophic risk, "including establishing agreed-upon standards, model testing and evaluation, compliance verification, third-party auditing, capability pacing, transparency in pre-deployment evaluation, incident notification, and any other risk management measures prioritized by the Secretary";
- focus on catastrophic risk, "namely (a) alignment and loss of control, (b) chemical, biological, radiological, nuclear weapons (CBRN), and (c) cybersecurity";
- negotiate "with the objective of reaching a treaty or other binding agreement between both nations to regulate frontier AI developers in each country to adhere to shared standards to address these risks, with clear mechanisms to verify compliance";
- begin bilaterally but "enable subsequent expansion of commitments based upon US-China agreements, including the establishment of an international AI safety body"; and
- establish a hotline between the two governments, "modeled on the 'red phone' between Washington and Moscow, for direct communication on AI risk."
Section 2 — CAISI initiative. CAISI "shall establish an initiative composed of AI safety experts from industry, academia, and civil society to engage with their counterparts in other countries to promote development and adoption of standards related to AI safety." In its work with foreign AI safety institutes, or through U.S. participation in international standards bodies "such as the International Organization for Standardization," CAISI is to prioritize the development and harmonization of AI safety testing and evaluation metrics, means to verify compliance, and other efforts needed to reach the bilateral measures in Section 1. "An amount of $100 million shall be authorized to support the operations of CAISI in this endeavor."
Section 3 — Safe harbor for expert communication. Until a treaty or other binding agreement under Section 1 is concluded, "AI industry and safety experts from universities, think tanks, and industry AI labs may communicate directly with peer experts within the People's Republic of China" without violating the following laws, which the outline lists: the Sherman Act §1 and other antitrust laws; the Economic Espionage Act (18 U.S.C. §§1831–1832); "the Espionage Act provisions"; 18 U.S.C. §951; the Foreign Agents Registration Act; the International Emergency Economic Powers Act; the Export Control Reform Act; and the Export Administration Regulations. The protection applies only if all of three conditions are met:
- Registration and U.S. participation. U.S. participants register with the Secretary of State "no fewer than 14 days prior to any such communication," and the Secretary may designate a member of government "who must be allowed to observe any communication, including electronic and verbal communications."
- Subject matter. Communication is "strictly limited to developing common understanding regarding testing and evaluation standards, protocols, metrics, and methods" needed for a U.S.–China agreement on catastrophic AI safety risks.
- Limitations. U.S. persons may not provide information that "(a) facilitates the distillation of U.S. large language models; (b) enables the release of intellectual property possessed by a U.S. person or entity, (c) shares non-public information about U.S. cybersecurity or national security vulnerabilities that could facilitate exploitation of those vulnerabilities."
Section 4 — Reporting. The Secretaries of State and Commerce report to Congress "with monthly updates on their progress."
Section 5 — Distillation negotiations. "Separately," the Secretary of Commerce, in consultation with other agencies, "shall seek to commence negotiations with the government of China regarding measures to ensure protection of U.S. intellectual property compromised by model distillation." See Adversarial Distillation.
Comparison with the press release
The press release of September 22, 2026 summarizes the same framework in seven points and adds statements by Liccardo. Its list of negotiation topics ("model testing and evaluation, independent auditing, transparency, incident reporting, capability pacing, and mechanisms to verify compliance") paraphrases Section 1, which reads "third-party auditing," "transparency in pre-deployment evaluation" and "incident notification." It describes the hotline as being "for urgent communication involving significant AI risks or incidents," where the outline says "for direct communication on AI risk." The release does not mention the possible international AI safety body, the list of statutes in Section 3, or the 14-day registration period, which it describes only as "advance registration with the State Department." Liccardo is quoted: "A lot of technical work must begin now, before political leadership can ever fruitfully meet at the negotiating table," "By establishing red lines now, we can avoid the need for red phones later," and "We don't talk to the Chinese because we trust them or agree with them; we shouldn't and don't. We talk to the Chinese because—as with the USSR during the Cold War—all of our lives depend on it" (Source: liccardo.house.gov).
Reception
Punchbowl News reported on September 22, 2026 that the framework was being laid out days before President Trump was to host Chinese President Xi Jinping for a summit, and noted that Treasury Secretary Scott Bessent had said the previous Sunday that the United States had proposed a "notification mechanism" for certain serious risks (Source: liccardo.house.gov, reproducing punchbowl.news). An IANS wire report dated September 23 summarized the proposal from the press release (Source: telanganatoday.com). The administration's parallel talks on an AI dialogue and incident line are described at US-China AI Competition: Different Races, Different Metrics.
Provenance
Captured on September 25, 2026 from the PDF on liccardo.house.gov (file name international-ai-cooperation-one-pager-v3-watermarked), the same URL the press release links. The text layer renders the capital I in "AI" as a lower-case l ("Al"); quotations on this page restore "AI." Punchbowl News hosted a copy under the same file name. The page date is the date of the press release that published the outline. No bill number or introduced text had been located as of September 25, 2026; when the text appears on congress.gov it becomes the primary text and this outline its companion.
Relationships
- supports: Liccardo–Kiley U.S.–China frontier AI safety negotiations bill (2026) — the outline of the bill.
- related: Sam Liccardo, Kevin Kiley — sponsors.
- depends-on: Pacing the Frontier (statement from employees of frontier AI companies, July 2026) — the employee statement the rationale cites.
- related: US-China AI Competition: Different Races, Different Metrics, Coordinated slowdown proposals compared — U.S.–China AI talks and pacing proposals in September 2026.
- related: NIST CAISI (Center for AI Standards and Innovation), International AI Standards — Section 2 initiative.
- related: Adversarial Distillation, Distillation — Sections 3 and 5.
- related: AI Antitrust — Section 3 includes a Sherman Act §1 safe harbor.
- related: CBRN Uplift, AI Incident Reporting, Open-Weight Frontier Models.