Multilateral treaty process engagement on AI

Movement organisations enter existing multilateral negotiation processes — the Council of Europe's AI treaty (2024), the UN Cybercrime Convention negotiations, UNESCO's Recommendation on the Ethics of AI, OECD AI Principles, the Global Partnership on AI (GPAI), the UN Advisory Body on AI, the ITU AI-for-Good process — as observers, civil-society participants, and coalition partners of sympathetic state delegations. The work is drafting-committee-level: attending negotiation rounds in Geneva, Strasbourg, Vienna, Paris and New York; submitting written comments through observer channels; briefing delegates from friendly member states before and between negotiation sessions; publishing scorecards of member-state positions; and coordinating across the civil-society coalition to speak with a shared voice when the negotiation asks for one. The output is treaty text: a specific clause the movement got in, a specific carve-out the movement kept out.

An actor chooses this strategy because multilateral instruments — even non-binding ones — anchor the normative floor on which subsequent national legislation, regulatory guidance, and jurisprudence stand: the Council of Europe AI Convention becomes the reference EU member states cite when translating the EU AI Act into national law; UNESCO's Ethics Recommendation becomes what national ministries cite when adopting AI frameworks; OECD AI Principles become the definitions national statutes borrow. The negotiations themselves take place on a schedule the movement can plan against, and observer status — earned once — provides recurring access to the room where the drafting happens. The strategy also converts the movement's cross-border coalition into a coordinated presence that a single national delegation cannot replicate, and it does so on terms (technical civil-society input, human-rights-framework grounding) member-state delegations often welcome as diplomatic cover for positions their governments hold but need external support to advocate for.

It trades enforceability for norm-setting reach. Multilateral instruments are typically non-binding or minimally enforceable — the Council of Europe AI Convention entered force with narrow ratifications and no meaningful enforcement mechanism, UNESCO recommendations do not bind member states, OECD principles create no rights of action — and even a well-drafted treaty text can be gutted at ratification, watered down at national implementation, or ignored in the absence of a monitoring body with real authority. The strategy is also resource-intensive per outcome: multilateral processes run over years, require permanent staff presence at the geographic host, and generate volumes of technical detail work whose only visible outputs are paragraphs of treaty text that most of the movement's supporters will never read. And civil-society-observer status is granted by the process's own governance and can be revoked or narrowed under state-delegation pressure — a permanent institutional relationship the movement can lose.

Ecology

Distinct from humanitarian disarmament treaty. Humanitarian-disarmament aims to create a specific new binding treaty banning a class of AI weapon (autonomous lethal weapons), on the model of the Mine Ban Treaty and the Convention on Cluster Munitions. This strategy engages the existing multilateral treaty-and-instrument ecosystem — the Council of Europe AI treaty already drafted, the UNESCO Recommendation already adopted, the GPAI process already running — to shape their output rather than launch a new instrument. The two are complementary registers of international governance work: create new binding frameworks where a specific harm justifies it; engage every existing framework where AI is being defined and constrained.

Distinct from international human rights body engagement. HR-body engagement works the UN human-rights machinery — Special Rapporteurs, Universal Periodic Review, treaty bodies — to produce authoritative human-rights findings and recommendations addressed to states. This strategy works the AI-specific multilateral treaty processes producing AI-specific normative instruments. The two feed each other — HR-body reports become citations in the treaty process, treaty text becomes reference material for HR-body findings — but their institutional venues, negotiation cycles, and civil-society entry paths differ.

Distinct from coalition lobbying of binding regional regulation. Regional-regulation lobbying targets a specific legislature (EU Parliament and Council on the AI Act, US Congress, national parliaments) with rights of amendment and a legislative timeline. This strategy targets multilateral negotiation bodies whose output is a treaty for signature and ratification — a different institutional shape, a different negotiation logic, and a different enforcement path. A coalition doing both operates across the national and multilateral layers of the same overall regulatory architecture.

Distinct from civil-society inside technical standards bodies. Technical standards bodies (ISO, IEEE, NIST) produce technical specifications; multilateral treaty processes produce normative and legal instruments. The distinction between "how the technology should be built" and "how the technology should be governed" runs through the movement's international work and produces separate coalitions with overlapping members and complementary outputs.