The Treaty Clause of Article II is one of the highest bars in American law. It was designed by the framers to force broad consensus on the nation's international commitments. In practice, in a polarised Senate, it means that a determined minority of 34 senators can block any treaty, regardless of its merits or the president's support. This is not a hypothetical constraint — it is the reason several major, carefully negotiated treaties never entered into force for the United States.
A history of treaties killed at home
- The Comprehensive Nuclear-Test-Ban Treaty (1999). Signed by the US and central to non-proliferation, it was rejected by the Senate 48–51, far short of 67. The US has still never ratified it.
- The UN Convention on the Law of the Sea. Supported by the military, industry, and successive presidents of both parties, it has sat unratified for decades because it cannot reach 67 votes.
- The Kyoto Protocol. The Senate pre-emptively voted 95–0 for a resolution opposing it, and it was never even submitted for ratification.
- The Convention on the Rights of Persons with Disabilities (2012). Rejected despite bipartisan backing and modelled on existing US law, falling five votes short.
The pattern is clear: even treaties with broad expert support, presidential backing, and allied pressure routinely fail to clear the Senate. An AI treaty — likely to be attacked as a surrender of sovereignty, a constraint on a strategic industry, and a concession to China — would face at least as steep a climb.
Why an AI treaty is especially vulnerable
The specific objections write themselves. Opponents will argue that binding limits hand an advantage to adversaries who will cheat; that verification amounts to letting foreigners inspect American labs; that the treaty smothers a critical industry; and that it cedes decisions about national security to an international body. Each of these arguments has sunk treaties before, and each maps directly onto the AI debate. A treaty that requires 67 votes in this environment is, on its face, a daunting proposition.
The paths around the two-thirds bar
Congressional-executive agreements
Not every binding international commitment is an Article II treaty. Many — including most US trade agreements — are approved by a simple majority of both houses as congressional-executive agreements. Courts have treated these as constitutionally interchangeable with treaties for most purposes. This route needs 51 Senate votes plus the House, not 67.
Sole executive agreements
The president can make certain binding commitments without Congress at all, within the scope of executive power. These are legally limited and can be undone by a successor, but they can carry initial commitments while broader ratification is pursued — the mechanism behind US adherence to the Paris Agreement.
Domestic legislation mirroring the treaty
Congress can enact the substance of an international standard as ordinary law by simple majority, binding US actors to the same obligations without formal ratification of the treaty itself. Compute and export controls already show this route in action.
Build durable bipartisan salience
The only path to actual ratification is making AI risk a genuinely cross-partisan concern, so that 67 votes becomes conceivable. That is a political project — the work of shifting the issue from partisan to consensus — that has to begin years before any treaty reaches the floor.
A treaty that cannot pass the Senate is a press release. Serious AI diplomacy has to be designed backward from 67 votes — or designed to use the binding instruments that do not need them.
Naoto Nakada, Founder · Nakada Foundation to Save Humanity
Designing for the domestic constraint
The lesson is not despair but design. The two-thirds bar is real, but it is not the only door. A well-constructed AI governance strategy would combine instruments: domestic legislation and executive action to establish immediate commitments, congressional-executive agreements where a simple majority is achievable, and a long-term political effort to build the bipartisan consensus that formal ratification would require. It would also shape the treaty's content to blunt the predictable objections — reciprocal verification rather than one-way inspection, protections for legitimate industry, and safeguards for national security. The mistake is to negotiate a treaty abroad while ignoring the arithmetic at home. In the American system, the hardest negotiation is often the one in the Senate, and it has to be planned for from the first day.
Common questions.
The Treaty Clause of Article II of the US Constitution requires the 'advice and consent' of two-thirds of the Senate — 67 of 100 votes — to ratify a treaty. The framers set the bar high to force broad national consensus on international commitments. In a polarised Senate, it means 34 senators can block any treaty regardless of its merits, which is why several major treaties have failed.
The Comprehensive Nuclear-Test-Ban Treaty was rejected 48–51 in 1999 and remains unratified. The UN Convention on the Law of the Sea has gone unratified for decades despite broad support. The Kyoto Protocol was pre-emptively opposed 95–0 and never submitted. The Convention on the Rights of Persons with Disabilities was rejected in 2012 despite bipartisan backing. Even well-supported treaties routinely fail to reach 67 votes.
Yes. Congressional-executive agreements, approved by a simple majority of both houses, are treated as largely interchangeable with treaties and cover most US trade deals. Sole executive agreements let the president make certain commitments alone, as with the Paris Agreement. Congress can also enact a treaty's substance as ordinary domestic law by simple majority. These routes need fewer votes but have their own legal and durability limits.
Because it would attract exactly the objections that have sunk treaties before: that it hands adversaries an advantage while they cheat, that verification means letting foreigners inspect American labs, that it smothers a strategic industry, and that it cedes national-security decisions to an international body. Overcoming these requires both careful treaty design and a years-long effort to make AI risk a genuinely bipartisan concern.