Article 01
Purpose
Establish a binding international framework that keeps the development and deployment of advanced AI systems aligned with human rights, democratic accountability, and long-term civilizational stability.
File 010 · Treaty
A draft Treaty for the age of intelligence.
A starting outline, not a finished document. Intended to be argued over, improved, and adopted in spirit by jurisdictions willing to set a floor.
Article 02
Definitions
Defines the terms used throughout the treaty: frontier AI system, general-purpose model, autonomous agent, training compute, deployment, developer, operator, and competent authority. Shared definitions are the floor for shared enforcement.
Article 03
Frontier AI Systems
Any AI system trained above a defined compute threshold, or demonstrating capabilities meaningfully beyond the current public state of the art, is classified as a frontier system and subject to the heightened obligations in this treaty.
Article 04
AGI Thresholds
Establishes objective indicators (capability evaluations, autonomy measures, resource acquisition tests) that trigger AGI-level obligations. Crossing a threshold triggers mandatory disclosure, third-party evaluation, and licensing review before further scaling or deployment.
Article 05
Compute Governance
Large-scale AI training compute is treated as a regulated input. Data centers above a defined capacity register with national authorities, report large training runs, and implement know-your-customer requirements for frontier customers.
Article 06
International Licensing
Frontier developers must hold a license issued by their national authority and recognised under this treaty. Licenses require demonstrated safety practices, incident reporting, and ongoing audit access.
Article 07
Independent Safety Audits
Before deployment and at regular intervals after, frontier systems undergo independent technical audits covering capability evaluations, misuse potential, alignment behaviour, and security posture. Audit results are shared with the treaty body.
Article 08
Model Evaluation Standards
An international standards body, with public participation, maintains evaluation suites for dangerous capabilities (cyber, biological, autonomous replication, deception, persuasion). Evaluations are versioned, reproducible, and publicly documented.
Article 09
Incident Reporting
Significant safety incidents, misuse events, and near-misses involving frontier systems must be reported to the national authority within a defined window. Aggregated, anonymised incident data is published.
Article 10
Emergency Containment
Each party maintains the legal and technical capacity to suspend training, restrict deployment, or require rollback of a frontier system that presents an imminent serious risk. Containment powers are subject to judicial review.
Article 11
Autonomous Weapons
Prohibits the development, deployment, and transfer of fully autonomous weapons systems that select and engage human targets without meaningful human control. Requires meaningful human control over all uses of lethal force.
Article 12
Synthetic Media
Requires provenance signals for AI-generated audio, video, and images at the point of creation. Platforms above a defined size must surface provenance information and provide accessible labelling for end users.
Article 13
Human Rights Protections
AI systems must not be used to enable mass surveillance, social scoring, predictive policing of protected groups, or other practices incompatible with international human rights law. Independent remedies must be available to affected persons.
Article 14
Labor and Economic Disruption
Parties commit to monitoring labor-market effects of advanced AI, funding transition support, protecting collective bargaining rights, and ensuring that productivity gains from AI are not captured solely by capital.
Article 15
Democratic Oversight
Frontier AI policy decisions, including licensing, evaluation standards, and emergency actions, are subject to legislative oversight, public consultation, and judicial review in each party state.
Article 16
Enforcement
Non-compliance can trigger graduated measures: fines proportionate to global revenue, suspension of licensing, restrictions on compute access, and, in serious cases, trade measures coordinated through the treaty body.
Article 17
Transparency
Parties publish annual reports covering licensed developers, audit summaries, enforcement actions, and incident statistics. Trade-secret carve-outs are narrowly defined and subject to independent review.
Article 18
International Coordination
Establishes a treaty body responsible for evaluation standards, mutual recognition of audits, information sharing on incidents, and coordinated response to cross-border risks.
Article 19
Public Accountability
Civil society, researchers, and affected communities have standing to submit evidence, request investigations, and participate in standards-setting. Whistleblower protections apply to staff at developers, auditors, and authorities.
Article 20
Human Primacy Clause
No provision of this treaty, and no decision taken under it, may be interpreted to subordinate human judgement, human rights, or democratic self-governance to the operation of any AI system.
No artificial intelligence system should be deployed at a scale capable of materially altering human civilization without enforceable public oversight, independent safety verification, and democratic accountability.