The governance literature on AI safety focuses heavily on the international negotiation problem: how do you get the US, China, the EU, and other major actors to agree on binding commitments? This is a genuine and hard problem. It receives less attention than it deserves, but it is not the only problem. The domestic ratification problem receives almost none. In the history of arms control, the domestic ratification stage has killed more treaties than failed international negotiations.
Anyone designing an AI governance strategy needs to plan for the domestic political fight, not just the international one.
How ratification works and why it is hard
International treaties become binding domestic law through ratification, which requires approval through each country's constitutional process. In the United States, that means a two-thirds Senate vote — 67 votes in the current chamber. This threshold, set by Article II of the Constitution, was designed to ensure that treaties reflected broad bipartisan consensus before becoming binding law. In practice, it gives the minority party the ability to block any treaty that lacks strong cross-partisan support, and it gives organized industries a small number of legislators to persuade in order to prevent ratification.
Other democracies have different ratification requirements, but all face versions of the same problem. Parliamentary majorities can change between when a government negotiates a treaty and when it seeks ratification. Domestic constituencies affected by the treaty's obligations have strong incentives to organize against it. Industries that perceive themselves as disadvantaged by the treaty will spend substantially to prevent ratification.
The history of US treaty ratification is, in significant part, a history of treaties signed and not ratified. SALT II, the CTBT, the Law of the Sea Convention (which the US signed but has never ratified, despite being a major maritime power), and the Convention on the Rights of the Child (ratified by every UN member state except the United States) are all treaties whose international negotiation succeeded and whose domestic ratification did not.
The SALT II case
SALT II illustrates how an agreement that took seven years to negotiate, involved the personal commitment of two heads of state, and had genuine strategic value for both sides could nonetheless fail at ratification because of domestic political conditions that shifted after the agreement was reached.
Carter submitted SALT II to the Senate in June 1979. Senate debate revealed significant Republican opposition centered on three arguments: that the treaty's verification provisions were inadequate; that the treaty gave the Soviet Union advantages in heavy missiles; and that arms control itself reflected a naive view of Soviet intentions. These arguments had been available throughout the negotiating period and had been addressed to various degrees in the treaty's text. They became politically decisive when the Soviet invasion of Afghanistan in December 1979 shifted the political environment. Carter withdrew the treaty from Senate consideration in January 1980, and it never came to a vote.
Both sides subsequently observed most of SALT II's limits voluntarily for several years, but the treaty never entered into legal force. The political window for ratification closed before the Senate voted, and it did not reopen.
The CTBT case
The Comprehensive Nuclear-Test-Ban Treaty was opened for signature in 1996 after years of negotiation. President Clinton signed it in September 1996. In October 1999, the Senate voted 51-48 against ratification, the first time the Senate had rejected a major international security treaty since the Treaty of Versailles in 1920.
The opposition arguments centered on verifiability and military necessity: senators who voted against ratification argued that the monitoring system could not reliably detect all nuclear tests above a certain threshold, and that the US needed the option to test nuclear weapons to maintain its arsenal's safety and reliability. The Clinton administration had not built sufficient political infrastructure for the ratification campaign, was caught off guard by the Senate majority leader's decision to schedule a vote without normal floor debate procedures, and lost the vote decisively.
The treaty has been in legal limbo since. The United States remains a signatory but not a party, bound only by the good-faith obligation of the Vienna Convention not to act against the treaty's object and purpose. Thirty years after it opened for signature, the CTBT has not entered into force because it requires ratification by eight specific countries — including the United States, China, India, and Pakistan — that have not ratified.
What AI industry opposition would look like
An AI safety treaty that required the United States to accept binding constraints on frontier AI development would face organized opposition from the AI industry and from national security constituencies who would argue that constraints disadvantage the United States relative to China.
The competitive disadvantage argument would be the most politically potent. The argument would run: a treaty that binds US AI companies to safety evaluation requirements and development constraints that Chinese companies do not observe functions as a unilateral restriction on US capability. This argument would be made regardless of whether the treaty actually bound China, because the political effectiveness of the argument does not depend on its accuracy. The Senate votes against SALT II were cast despite the fact that the treaty bound both sides symmetrically.
The economic harm argument would accompany it. AI companies generating significant revenue and employing tens of thousands of high-wage workers would present data on the economic consequences of development constraints. Legislators in states with significant AI industry presence would face direct constituency pressure to oppose ratification.
The verifiability argument, which proved decisive for the CTBT, would reappear for AI governance. Any verification mechanism that requires disclosure of AI training data, model architecture information, or hardware configurations would face opposition from companies concerned about competitive exposure and from national security agencies concerned about classified AI programs. The argument that the verification system is inadequate to detect sophisticated violations while imposing burdens on compliant actors is structurally available for any verification design.
What successful ratification campaigns have done
Treaties that have successfully navigated US Senate ratification share several features that distinguish them from treaties that did not.
They built bipartisan support in the Senate before the floor vote, not during it. The Chemical Weapons Convention, ratified in 1997 by a vote of 74-26, had sustained Republican support because Republican senators were convinced that the alternative — the US outside a near-universal treaty banning chemical weapons — was a worse outcome than the treaty's imperfections. The administration worked individual senators for months before the floor vote and had counted their votes before asking for debate to begin.
They mobilized constituencies outside government in support. The CWC ratification campaign included active support from the American Chemical Society, major US chemical companies who concluded that treaty participation was in their commercial interest, and veterans' organizations who supported the chemical weapons ban on humanitarian grounds. The campaign against the treaty was largely confined to ideological anti-treaty voices without comparable institutional weight.
They had presidential attention sustained through ratification, not just through signing. Carter's withdrawal of attention from SALT II ratification as his presidency became consumed by Iran and Afghanistan was a contributing factor in the treaty's failure. The CWC ratification received sustained White House engagement, including personal lobbying by President Clinton and senior national security officials.
"An AI safety treaty that reaches the Senate without years of domestic coalition-building behind it will face the same fate as SALT II. The political work has to start long before the treaty is signed."
Naoto Nakada, Founder · Nakada Foundation to Save Humanity
For AI governance, the domestic ratification challenge is not separate from the international negotiation challenge. They are the same challenge approached from different angles. A treaty text that is drafted with US Senate ratification in mind — that addresses the verifiability objection, that includes provisions making the competitive disadvantage argument harder to sustain, and that is designed to attract bipartisan support — is more likely to succeed internationally because it reflects the constraints of the most demanding ratification environment. And a domestic coalition-building campaign that is underway before the treaty is concluded gives the administration credibility in international negotiations that it lacks if it is negotiating a treaty it cannot commit to ratifying.
Common questions.
Under Article II of the Constitution, the President negotiates and signs treaties, but ratification requires approval by two-thirds of the Senate — currently 67 votes. This high threshold gives the minority party significant blocking power over any treaty lacking strong cross-partisan support. Several major arms control treaties have been signed but not ratified because the Senate declined to act or voted against them, including SALT II and the Comprehensive Nuclear-Test-Ban Treaty.
SALT II was signed in June 1979 after seven years of negotiations. President Carter withdrew it from Senate consideration in January 1980 after the Soviet invasion of Afghanistan made ratification politically impossible. The Senate had already expressed concern about the treaty's verification provisions; the geopolitical shift made the votes unavailable. Both sides observed most of its limits voluntarily for several years, but it never entered into legal force — the canonical example of a successfully negotiated arms control agreement that failed at domestic ratification.
Based on patterns from other industries that have faced treaty-based regulation, AI industry opposition would focus on competitive disadvantage arguments (governance requirements would constrain US companies more than Chinese ones), economic harm arguments, technical infeasibility of verification mechanisms, and national security arguments about AI capability constraints affecting the US military. These arguments would be made through industry associations, direct lobbying, and campaign contributions to legislators with AI policy jurisdiction.
Several factors matter. Treaties with sustained executive branch advocacy through the full ratification period fared better. Treaties with clear public benefits were more defensible than those with abstract strategic justifications. Treaties where the opposition could be framed as leaving the country exposed to the treaty's subject matter did better than those where opposition could be framed as protecting national capability. And treaties ratified during periods of bipartisan foreign policy agreement fared better than those caught in partisan polarization — a condition that directly affects any near-term AI governance ratification effort in the United States.