Governance
Principles only matter if they are upheld.
The Accord is backed by a practical, layered governance model — designed to be adopted by nations and institutions, resistant to capture, and transparent to the public it serves.
The model
Four layers that keep intelligence accountable.
No single authority can be trusted with power this large, and none should be. Accountability is held instead across four interlocking layers — each a check on the others, none able to capture the whole.
The Horizon Council
A standing, multi-stakeholder body — scientists, ethicists, civil society, and signatory representatives — that stewards the Accord and interprets it as technology evolves. It speaks for no single nation or company, and answers to the public it serves.
Conformance & Audit
Independent assessors evaluate high-impact systems against the Accord. Findings are published; serious breaches trigger remediation and public disclosure.
Tiered Obligations
Requirements scale with capability and risk. A creative tool and a system governing critical infrastructure are not held to the same bar — but neither is exempt.
Open Revision
The Accord is amended through a transparent, deliberative process. Proposed changes are debated in public and ratified by signatories.
Who does what
Shared responsibility, clearly drawn.
Accountability works only when everyone knows their part. The Accord assigns clear duties across the chain — from the people who build systems to the institutions that govern them.
Builders
Design for oversight, document capabilities and limits, disclose AI involvement, and submit high-impact systems for review.
Operators
Maintain human-in-the-loop controls, monitor real-world impact, and provide channels to contest and reverse decisions.
Signatory nations
Adopt the Accord into policy, fund independent oversight, and cooperate across borders rather than racing to the bottom.
The Council
Interpret the Accord, coordinate audits, publish findings, and shepherd open revision in the public interest.