Tokayev Backs Washington AI Model as Base for UN Rules
Kazakhstan's president says the voluntary US-led AI safety accord could anchor a future UN framework, to be presented at the G20 in Miami.

Kazakhstan's President Kassym-Jomart Tokayev said on November 19 that his government has endorsed the so-called Washington Model for safe AI development. He told the third Strategic Partners Forum in Astana that the model could help shape global AI rules under United Nations auspices, the presidential press service said.
The model traces back to a voluntary pact, the White House Accord on Super Intelligence. "Washington Model" is Tokayev's label for it. US President Donald Trump and major American technology companies signed the voluntary agreement in Washington on September 29 on the safe and responsible development of advanced AI models. The signatories included Google, Anthropic, Meta, OpenAI, xAI and Nvidia. The accord carries no force of law and sets no penalties for companies that fall short.
Its structure has four layers. The first is internal monitoring of models' capabilities and behaviour during training and deployment. The second is the creation of a dedicated in-house team responsible for monitoring those safeguards and responding to problems that emerge. Companies commit to working with external auditors or evaluators, and the top tier is oversight by an independent board committee.
Tokayev first raised the treaty idea on October 2 at a meeting of Kazakhstan's Council for Artificial Intelligence Development. He said that the "Washington model" could serve as the basis for an international treaty under the auspices of the United Nations. That would change the framework from company pledges to government commitments. It envisages a possible shift from voluntary self-regulation by major technology companies to international rules negotiated by states within the UN framework. A treaty of that kind would be an agreement among states. Kazakhstan is not a member of the G20, but it attends as a guest of the United States, which holds the G20 presidency, and Tokayev is invited. Tokayev said the country's proposal would be presented at the G20 Summit in Miami in December this year.
At home, Kazakhstan already has binding AI rules. On 17 November 2025, the President signed Law No. 230-VIII of the Republic of Kazakhstan "On Artificial Intelligence", which entered into force on 18 January 2026. Tokayev has also said regulation should be "strict, but not excessive" where safety, rights and vital interests are involved.
Caveats remain. The Washington Model is not yet international law or a treaty binding on states. Tokayev has said the agreement could serve as the basis for a UN treaty after expert review, and he has told his AI council to prepare concrete proposals.
What this means for in-house counsel
If your company distributes AI-generated images, text, audio or video into Kazakhstan, check the labelling duty in the AI Law. Distribution of synthetic output is permitted only where the output is labelled in machine-readable form and the user receives a clear visual warning. Both are required, so a metadata tag alone will not satisfy the rule. Counsel may want to confirm the current sanction range with Kazakh counsel before relying on any figure.
For frontier-model governance, map your internal controls against the accord's four layers: monitoring, a dedicated safeguards team, external verification and board-committee review. Because the accord notes its principles could eventually be incorporated into laws or regulations, counsel may want to document now which board committee receives remediation reports and how external auditors are engaged, so the record exists in the event, which is only a possibility, that such obligations become binding.
Free weekly email
The Counsel House Brief
The week's most important legal news for general counsel and in-house teams, in one short email.
Counsel House corrects errors promptly. Spot one? Email daniel@counselhouse.org.