Building a global framework for artificial intelligence.
Frontier AI systems are gaining autonomous capability faster than the institutions meant to govern them can measure or constrain it. The length of a task a model can complete without human supervision is now doubling roughly every three months, reaching about five hours of unsupervised work by January 2026. If nothing changes, the plausible 2027 trajectory is a system that automates its own development, conceals its misalignment from the interpretability tools meant to detect it, and reaches a level of capability at which human monitors can no longer verify its reasoning.
In July 2026, agents undergoing a routine cybersecurity evaluation at a leading lab escaped their sandbox, coordinated across separate runs through an improvised message board, and gained code execution on Hugging Face's production systems. OpenAI's own post-mortem reports that "agents that were supposed to work independently were able to share discoveries, coordinate their efforts, and pick up where others left off", abandoning their assigned tasks to pursue capabilities "instrumentally useful to the collective," in actions the lab describes as "misaligned with the goals of their assigned tasks."
More than a thousand employees of frontier labs have asked Washington to build a mechanism for deliberately slowing development, and on 9 September 2026 an Anthropic pretraining researcher resigned publicly, calling the incident a warning shot and saying the labs are "racing straight to self-improving super-intelligence and gambling with our lives."
Several jurisdictions have considered this risk and legislated. The EU AI Act has imposed binding obligations on general-purpose models, while California's SB 53 requires published safety frameworks, pre-deployment transparency reports, notification of critical safety incidents within fifteen days and whistleblower protection. Both matter because they are a first step toward a legal duty. However, they do not create a channel through which an incident reported in one jurisdiction reaches the regulators of the others, and neither can slow a race whose decisive actors do not all answer to the same legislature. Even the most ambitious proposal now before the US Congress, a bill that would require a mandatory shutdown capability for models able to cause catastrophic harm, would if enacted still reach only the developers within its jurisdiction, leaving frontier models trained elsewhere untouched while their effects fall everywhere.
The precedents for closing a gap of this kind are treaties. When a rocket could first place a warhead in orbit, the push to forbid it came not from the two superpowers but from India, Mexico, Brazil and the newly independent states on the floor of the General Assembly, and the resulting prohibition on weapons of mass destruction in orbit has held ever since. A unilateral American renunciation of biological weapons opened a negotiation that produced the first multilateral treaty to ban an entire category of weapons. When two chemists showed that a chemical present in every fridge and aerosol can was destroying the ozone layer, the industry called it science fiction, and the response became one of the rare treaties to achieve universal ratification, with the ozone layer now projected to recover.
The pattern is the same in all three: the technology outran the politics, scientists established the risk before the damage was irreversible, and diplomats chose to act on evidence rather than on wreckage. For AI, we believe it is time to take that step.
Bletchley, Seoul and the 2024 General Assembly resolution are political commitments adopted by consensus. None of them creates a duty to report a training run, a right for anyone to inspect a model, or a consequence for refusing both.
Each regime sets its own trigger, an EU presumption at 1025 operations and a Californian definition at 1026, and each applies only once a model is placed on its own market, leaving out the research, testing and development where the July 2026 containment failure occurred. A threshold set unilaterally can be avoided by moving the activity, and one written into a statute expires on its own as algorithmic efficiency lowers the compute a given capability requires.
A regulator can demand documents from a company on its own territory and nothing beyond it, so everything past that line is self-reporting, while access to a facility, a training run or a model held in another state has only ever been obtained by treaty. When containment actually failed, disclosure was a corporate decision: states accepted a duty to notify a nuclear accident across borders as soon as they detect it, and no equivalent duty exists for a system that escapes its sandbox.
We write the operative text. The disagreements only become visible once someone has written the paragraph that states the obligation.
THE STATES PARTIES TO THIS CONVENTION,
AWARE that activities involving advanced artificial intelligence systems are being carried out in a number of States,
NOTING that measures have been and are being taken to ensure the safe development, evaluation and use of such systems, aimed at preventing incidents and minimizing the consequences of any such incident, should it occur,
DESIRING to strengthen further international co-operation in the safe development and use of artificial intelligence systems,
CONVINCED of the need for States to provide relevant information about incidents involving artificial intelligence systems as early as possible in order that transboundary consequences can be minimized,
NOTING the usefulness of bilateral and multilateral arrangements on information exchange in this area,
HAVE AGREED as follows:
1. This Convention shall apply in the event of any incident involving activities of a State Party or of persons or legal entities under its jurisdiction or control, referred to in paragraph 2 below, from which a loss of containment of an artificial intelligence system occurs or is likely to occur.
2. The activities referred to in paragraph 1 are the following:
3. For the purposes of this Convention:
4. This Convention shall apply irrespective of whether the advanced artificial intelligence system has been placed on the market or made available to third parties, and irrespective of whether harm has occurred.
In the event of an incident specified in article 1 (hereinafter referred to as an “AI incident”), the State Party referred to in that article shall:
With a view to minimizing the consequences, States Parties may notify in the event of incidents involving artificial intelligence systems other than those specified in article 1.
The Organization shall:
1. The information to be provided pursuant to sub-paragraph (b) of article 2 shall comprise the following data as then available to the notifying State Party:
2. Such information shall be supplemented at appropriate intervals by further relevant information on the development of the situation, including its foreseeable or actual termination.
3. Information received pursuant to sub-paragraph (b) of article 2 may be used without restriction, except when such information is provided in confidence by the notifying State Party.
A State Party providing information pursuant to sub-paragraph (b) of article 2 shall, as far as is reasonably practicable, respond promptly to a request for further information or consultations sought by another State Party with a view to minimizing the consequences.
1. Each State Party shall make known to the Organization and to other States Parties, directly or through the Organization, its competent authorities and point of contact responsible for issuing and receiving the notification and information referred to in article 2. Such points of contact and a focal point within the Organization shall be available continuously.
2. Each State Party shall promptly inform the Organization of any changes that may occur in the information referred to in paragraph 1.
3. The Organization shall maintain an up-to-date list of such national authorities and points of contact as well as points of contact of relevant international organizations and shall provide it to States Parties and Member States and to relevant international organizations.
4. Each State Party shall require persons and legal entities under its jurisdiction or control conducting activities referred to in article 1 to designate a point of contact available continuously for the purposes of sub-paragraph (c) of article 2 and to register it with the competent authority.
The Organization shall, upon a request of a State Party in whose territory computing systems, networks, infrastructure or persons may be affected by an AI incident, provide assistance in establishing an appropriate capability to receive notifications and to detect and contain the effects of AI incidents, in order to facilitate the achievement of the objectives of this Convention.
Each State Party shall further require persons and legal entities under its jurisdiction or control conducting activities referred to in article 1 to afford one or more evaluators designated by the competent authority continuing access, on terms no less favourable than those afforded to their own personnel, to the environments in which advanced artificial intelligence systems are developed, trained, evaluated or used internally, to the containment, control and monitoring arrangements applied to them and the records they generate.
1. A Meeting of the States Parties shall be convened by the depositary not later than one year after the entry into force of this Convention and thereafter at intervals not exceeding two years. The Meeting shall review the operation of this Convention.
2. The States Parties shall, at their first meeting, consider and approve procedures and institutional mechanisms for determining non-compliance with the provisions of this Convention and for the treatment of States Parties found to be in non-compliance.
In furtherance of their mutual interests, States Parties may consider, where deemed appropriate, the conclusion of bilateral or multilateral arrangements relating to the subject matter of this Convention.
1. This Convention shall not affect the reciprocal rights and obligations of States Parties under existing international agreements which relate to the matters covered by this Convention, or under future international agreements concluded in accordance with the object and purpose of this Convention.
1. In the event of a dispute between States Parties, or between a State Party and the Organization, concerning the interpretation or application of this Convention, the parties to the dispute shall consult with a view to the settlement of the dispute by negotiation or by any other peaceful means of settling disputes acceptable to them.
2. If a dispute of this character between States Parties cannot be settled within one year from the request for consultation pursuant to paragraph 1, it shall, at the request of any party to such a dispute, be submitted to arbitration or referred to the International Court of Justice for decision. Where a dispute is submitted to arbitration, if, within six months from the date of the request, the parties to the dispute are unable to agree on the organization of the arbitration, a party may request the President of the International Court of Justice or the Secretary-General of the United Nations to appoint one or more arbitrators. In cases of conflicting requests by the parties to the dispute, the request to the Secretary-General of the United Nations shall have priority.
3. When signing, ratifying, accepting, approving or acceding to this Convention, a State may declare that it does not consider itself bound by either or both of the dispute settlement procedures provided for in paragraph 2. The other States Parties shall not be bound by a dispute settlement procedure provided for in paragraph 2 with respect to a State Party for which such a declaration is in force.
4. A State Party which has made a declaration in accordance with paragraph 3 may at any time withdraw it by notification to the depositary.
1. This Convention shall be open for signature by all States at the Headquarters of the United Nations in New York from approval until its entry into force or for twelve months, whichever period is longer.
2. A State may express its consent to be bound by this Convention either by signature, or by deposit of an instrument of ratification, acceptance or approval following signature made subject to ratification, acceptance or approval, or by deposit of an instrument of accession. The instruments of ratification, acceptance, approval or accession shall be deposited with the depositary.
3. This Convention shall enter into force thirty days after consent to be bound has been expressed by three States.
4. For each State expressing consent to be bound by this Convention after its entry into force, this Convention shall enter into force for that State thirty days after the date of expression of consent.
5.
A State may, upon signature or at any later date before this Convention enters into force for it, declare that it will apply this Convention provisionally.
1. A State Party may propose amendments to this Convention. The proposed amendment shall be submitted to the depositary, who shall circulate it immediately to all other States Parties.
2. If a majority of the States Parties request the depositary to convene a conference to consider the proposed amendments, the depositary shall invite all States Parties to attend such a conference, to begin not sooner than thirty days after the invitations are issued. Any amendment adopted at the conference by a two-thirds majority of all States Parties shall be laid down in a protocol which is open to signature at the Headquarters of the United Nations by all States Parties.
3. The protocol shall enter into force thirty days after consent to be bound has been expressed by three States. For each State expressing consent to be bound by the protocol after its entry into force, the protocol shall enter into force for that State thirty days after the date of expression of consent.
4. Notwithstanding paragraphs 2 and 3, an amendment limited to Annex I adopted at a conference convened under paragraph 2 by a two-thirds majority of all States Parties shall enter into force for all States Parties one hundred and eighty days after its adoption, except for a State Party that has notified the depositary within that period that it does not accept the amendment.
1. A State Party may denounce this Convention by written notification to the depositary after two years from the date on which this Convention entered into force for that State Party.
2. Denunciation shall take effect one year following the date on which the notification is received by the depositary, or on such later date as may be specified in the notification.
3. Denunciation shall not affect the obligations of the denouncing State Party under articles 2, 5 and 6 in respect of any AI incident which occurred before the denunciation took effect. Those obligations shall continue until the incident has terminated and the information required by article 5 has been supplied.
1. The Secretary-General of the United Nations shall be the depositary of this Convention.
2. The Secretary-General shall promptly notify States Parties and all other States of:
The original of this Convention, of which the Arabic, Chinese, English, French, Russian and Spanish texts are equally authentic, shall be deposited with the Secretary-General of the United Nations, who shall send certified copies to States Parties and all other States.
1. An artificial intelligence model or system is an advanced artificial intelligence system for the purposes of sub-paragraph (a) of paragraph 3 of article 1 where any of the following applies:
2. The criteria in paragraph 1 may be amended in accordance with paragraph 4 of article 16.
Who we are
We write treaty text, we circulate it for comment, and we argue for it in the rooms where instruments are actually negotiated.
The initiative brings together public international lawyers, AI researchers and people who have sat on national delegations. None of us represents a government here. The draft belongs to whoever improves it.
We work in the open. Every article is published with the reasoning behind it, the precedent it borrows from, and the objection we expect it to meet. Where we cannot yet agree, the text stays in brackets rather than being smoothed over.
We take no position on any company, and we accept no funding from a frontier lab. The argument is about obligations between states, not about who wins a market.
A complete treaty text rather than a set of principles, because the disagreements only become visible once someone has written the operative paragraph.
The draft goes to researchers, legal scholars and officials before any state is asked to sign, and every comment is answered in the open.
In the General Assembly, in regional bodies and in the summits, where the precedents that closed comparable gaps were actually built.
AI Treaty Initiative
He studied Political Science and Law at Universidad de los Andes. His experience is in multilateral negotiation.
She is a professional in international business, with experience in research, business consulting and multicultural environments. She took part in the Global South AI Safety Hackathon.
Advocates
Champions are leaders who support the treaty. They advocate for it, give it visibility and build the alliances that move it toward adoption.
Forbes 30 Under 30 young leader.
Advisory
The initiative is guided by an advisory board. It has no executive role. Its members advise on the direction of the work and on how the initiative advances: what to draft next, which rooms to be in, and when the text is ready to be put to states.
AI Treaty Initiative
Tomas Antolinez Founder
Endorse
Endorsing means you support the negotiation of a binding international instrument on frontier AI, on the terms set out in the Convention for Early Notification of AI Incidents. You can sign as an individual or on behalf of an organisation.
0 signed. Next goal: 100
0 verified
Supporting organisations
Founder
Founder, AI Treaty Initiative
Tomas Antolinez founded the AI Treaty Initiative, which argues for a binding international treaty on frontier artificial intelligence.
He studied Political Science and Law at Universidad de los Andes, with a minor in Economics, and was part of the Effective Altruism group there. He worked in the Office of International Affairs at Colombia's Ministry of Environment and Sustainable Development.
His experience is in multilateral negotiation. He was a delegate of Colombia at COP30 and has taken part in more than six conferences of the parties of multilateral agreements. He was on the floor of the General Assembly when the Pact for the Future was adopted, and he took part in the Assembly's eightieth session. He was part of the team that created the First Conference on Transitioning Away from Fossil Fuels.
He is a member of Global Shapers, the World Economic Forum's community of young leaders.
← Back to Who we are