Executive summary

Resources / Regulatory Frameworks

Council of Europe Framework Convention on AI

The first international legally binding treaty on AI, in force since November 2025, and the institutional answer to a question the EU AI Act could not address alone

Why the Convention exists, and what it is not

The most common framing of the Council of Europe Framework Convention on AI describes it as the international counterpart to the EU AI Act — the binding treaty that complements the regulation. The framing is convenient and structurally incomplete. The Convention was not built to translate the AI Act into international law. It was built to do something the AI Act, by its nature, cannot: extend a human-rights-based model of AI governance beyond the European Union, on terms that non-EU democracies can accept.

The dominant narrative reads the Convention through the EU AI Act lens: same continent, same year, same broad concerns about high-risk AI. That reading misses the institutional logic. The Council of Europe is not the European Union. It is a separate intergovernmental organisation with 46 member states whose mandate is human rights, democracy, and the rule of law — not the internal market. The Council of Europe operates through binding treaties that any state in the world can join, not through directives or regulations binding only on EU members. Its greatest historical contribution to digital policy is Convention 108 (1981), the first international binding treaty on data protection, which preceded the GDPR by 37 years and influenced data protection regimes from Latin America to Africa to Asia. The Framework Convention on AI is consciously the heir to that lineage.

What the Convention actually does, then, is not regulate AI at the level of conformity assessment, technical standards, or product placement. It establishes the rights-based perimeter within which any party — EU member, non-EU democracy, or any state that chooses to accede — agrees that AI development and deployment must operate. The substantive obligations are framed in terms of human rights protection, democratic process integrity, and rule of law preservation. The level of detail is deliberately lower than the AI Act’s. Higher detail would have made the treaty unsignable for non-EU parties whose legal systems differ structurally from the EU’s.

This shapes the institutional choice in three ways. First, the Convention is technology-neutral. Where the AI Act distinguishes between unacceptable, high-risk, limited-risk, and minimal-risk systems with detailed rules for each, the Convention articulates obligations that apply across the lifecycle regardless of risk classification. Second, the Convention defines obligations of result rather than obligations of means. Parties commit to outcomes (protection of rights, transparency, accountability, redress); they choose the legislative or administrative measures appropriate to their constitutional system. Third, the Convention treats the public sector and private sector asymmetrically. The substantive obligations bind public authorities and entities acting on their behalf directly. For private actors, each Party declares at signature how it intends to give effect to the obligations — through direct application of the principles or through other appropriate measures. This declaration mechanism is the structural compromise that allowed the United States and the United Kingdom, with regulatory traditions different from the EU’s, to sign without committing to AI Act–style horizontal regulation of private actors.

The Convention entered into force on 1 November 2025. It is the operational reference for an international AI rights regime that, like Convention 108 before it, is likely to outlive any single national regulation and to shape how courts, regulators, and litigants in adopting jurisdictions interpret the relationship between AI and fundamental rights.

Subjective and material scope

Who it addresses. The Convention addresses Parties — states (and the European Union) that have ratified, accepted, approved, or acceded to the treaty. Each Party undertakes obligations to give effect to the Convention’s provisions through domestic legal and administrative measures.

Negotiation involved 46 Council of Europe member states, the European Union, and 11 non-member states (Argentina, Australia, Canada, Costa Rica, Holy See, Israel, Japan, Mexico, Peru, the United States, Uruguay), with private sector, civil society, and academic representatives as observers. The Convention is open to accession by any state in the world, member of the Council of Europe or not, on terms specified in the treaty.

Initial signatories at opening (5 September 2024): Andorra, Georgia, Iceland, Norway, Republic of Moldova, San Marino, the United Kingdom, Israel, the United States, and the European Union. Subsequent signatures expanded the list. By early 2025, signatories numbered over thirty when counting EU member states represented through the Commission’s signature. Ratifications proceeded through 2025; the United Kingdom, France, and Norway were among the early ratifiers, with the threshold met to bring the Convention into force on 1 November 2025. The European Parliament gave consent to EU conclusion of the Convention in February 2026 (A10-0007/2026).

What it covers. Activities within the lifecycle of AI systems carried out by:

– Public authorities and persons acting on their behalf — directly bound by the substantive obligations.
– Private actors — bound through whatever legislative, administrative, or other measures each Party commits to in its declaration at signature or ratification.

The lifecycle scope means the Convention reaches design, development, deployment, use, and decommissioning of AI systems. Excluded are activities related to national security interests (Article 3(2)), provided those activities are carried out in a manner consistent with international human rights law and respect democratic institutions and processes. Research and development activities not yet in deployment are also outside the principal obligations, subject to caveats. Activities of natural persons for purely personal, non-professional purposes are excluded.

The definition of AI system used in the Convention follows the OECD definition, ensuring interoperability with the broader international regime that has crystallised around that definition.

The substantive obligations

The Convention is structured into eight chapters and 26 articles. The substantive content sits primarily in Chapters II through V, which articulate the principles and rights the Convention protects.

Chapter II — Fundamental principles for AI systems

Article 7 establishes human dignity and individual autonomy as foundational. AI systems must be designed, developed, used, and decommissioned in ways that respect human dignity and individual autonomy.

Article 8 addresses transparency and oversight. Each Party must adopt measures ensuring transparency and oversight of AI systems tailored to their specific contexts and risks, including identifying content generated by AI systems where appropriate.

Article 9 addresses accountability and responsibility. Each Party must adopt measures ensuring accountability and responsibility for adverse impacts on human rights, democracy, and the rule of law arising from AI systems.

Article 10 addresses equality and non-discrimination. AI systems must not undermine equality, including gender equality, or produce unjustified discrimination.

Article 11 addresses privacy and data protection. AI systems must respect privacy rights and personal data protection, with reference to existing instruments including Convention 108 and the modernised Convention 108+.

Article 12 addresses reliability. AI systems must be reliable and have adequate performance to deliver their intended purposes safely.

Article 13 addresses safe innovation. Parties must, where appropriate, enable controlled environments for the development, experimentation, and testing of AI systems under the supervision of competent authorities.

Chapter III — Remedies

Article 14 requires effective procedural guarantees, safeguards, and rights for persons affected by AI systems, including the ability to lodge complaints with competent authorities and to seek effective remedies for violations.

Article 15 requires that persons interacting with AI systems be notified that they are interacting with an AI system rather than a human, where appropriate. This mirrors Article 50 of the EU AI Act on transparency obligations for AI system providers and deployers.

Chapter IV — Assessment and mitigation of risks and adverse impacts

Article 16 establishes risk and impact management framework obligations. Parties must adopt measures for the identification, assessment, prevention, and mitigation of risks and adverse impacts of AI systems on human rights, democracy, and the rule of law. The measures must be graduated and differentiated according to severity and probability of impact. Parties must also assess the need for moratoriums, bans, or other measures concerning AI uses incompatible with human rights standards.

Chapter V — Implementation

Articles 17 to 22 cover implementation: non-discrimination obligations to vulnerable groups (Article 17), protection of digital literacy and skills (Article 18), public consultation requirements (Article 19), digital literacy and skills (Article 20), and the safeguarding of existing protections (Articles 21 and 22).

Chapter VI — Follow-up mechanism and co-operation

Articles 23 and 24 establish the Conference of the Parties, the institutional mechanism through which Parties monitor implementation, share experience, and consider necessary amendments. The Conference brings together Parties, observers, and external expertise to ensure the Convention adapts as AI technology evolves.

The remaining chapters address final clauses including the rules on signature, ratification, entry into force, accession, territorial application, federal state clauses, and amendments.

How the Convention interacts with the EU AI Act

The Convention and the EU AI Act are not redundant. They operate at different levels of the regulatory hierarchy.

The EU AI Act is the EU’s implementation of obligations broadly compatible with the Convention. The European Parliament made the relationship explicit in the consent procedure for EU conclusion of the Convention: the AI Act provides the EU’s mechanism for satisfying the Convention’s obligations within EU territory. This means EU member states do not need to adopt additional national measures to comply with the Convention’s substantive obligations as long as those obligations are satisfied through the AI Act.

For non-EU Parties, the picture is different. The United Kingdom’s pro-innovation regulatory approach, the United States’ patchwork of executive orders and sectoral regulation, Canada’s evolving AIDA framework, and other national systems must each find their own path to satisfying the Convention’s obligations of result. The declaration mechanism for private sector application means each Party can choose how it discharges the obligations — direct application of principles, sectoral regulation, contractual mechanisms, or combinations thereof. This flexibility is what makes the Convention signable across regulatory traditions that the AI Act could not bridge.

The practical implication is that the Convention will, over time, function as the rights-based floor beneath divergent national AI regimes. Where national systems differ substantially in their substantive rules, they will increasingly converge in the rights they protect because the Convention requires it. This is the same dynamic that Convention 108 produced in data protection: divergent national data protection laws, anchored to a common rights-based floor, with binding international scrutiny through the Convention’s monitoring mechanism.

Governance and follow-up

The Convention establishes the Conference of the Parties as its principal follow-up body. The Conference is convened by the Secretary General of the Council of Europe and meets to:

– Review implementation of the Convention by Parties.
– Make recommendations regarding interpretation and application.
– Consider amendments proposed by any Party.
– Promote international cooperation among Parties.
– Engage with non-Party observers, civil society, academia, and industry.

Monitoring is non-judicial. The Conference produces recommendations, peer reviews, and shared interpretive guidance. Parties remain responsible for domestic implementation and for the legal mechanisms that give effect to the Convention’s obligations within their territory.

The Convention is also linked to the broader Council of Europe legal architecture. The European Court of Human Rights, while not directly enforcing the Convention, is expected to draw on it when interpreting Article 8 (privacy) and other rights protected under the European Convention on Human Rights in cases involving AI systems. This indirect judicial reach has been the principal enforcement channel for Convention 108 over four decades and is likely to operate similarly for the AI Convention.

Intersections with other instruments

The Convention does not operate in isolation. Five intersections shape its operational role.

EU AI Act. Treated above. The AI Act is the EU’s mechanism for satisfying Convention obligations within EU territory. Convention obligations are conceptually narrower than AI Act obligations — the Convention is rights-focused, the AI Act is product-safety-and-rights-focused. Compliance with the AI Act will, by construction, satisfy most of what the Convention demands of EU Parties.

Convention 108 and 108+. Direct lineage. The 1981 Convention on the Protection of Individuals with regard to Automatic Processing of Personal Data and its 2018 modernised version are explicitly referenced in the AI Convention’s privacy and data protection provisions (Article 11). The institutional model — binding international treaty open to non-Council of Europe states, monitored through a Convention Committee, drawing rights-based floor across divergent national systems — is the same.

OECD AI Principles. Compatible. The Convention’s definition of AI system follows the OECD definition. The substantive obligations align with OECD values-based principles, although the Convention is binding where the OECD framework is not.

UNESCO Recommendation on the Ethics of AI. Compatible at the conceptual level. UNESCO’s broader values orientation flows into the Convention’s rights orientation, but the Convention is narrower (focused on rights, democracy, rule of law) and binding (UNESCO is not).

National frameworks of non-EU Parties. This is where the Convention will produce most of its operational effect. The United Kingdom’s upcoming AI legislation, the United States’ evolution of executive orders into sectoral and state-level regulation (including the Colorado AI Act), Canada’s AIDA, and other national frameworks must each find their domestic mechanism for satisfying Convention obligations. The mechanism varies; the obligation does not.

> ## ⚖️ How Zertia operates within the Council of Europe Framework Convention
>

>
>

> > ### Built for Council of Europe Convention compliance from day one
>

>
>

> Accreditations and memberships: 🎖️ ANAB-accredited (US) · 🎖️ UKAS process (UK) · 🎖️ ENAC process (EU) · 🏛️ IAPP member · 🏛️ INCITS member · 🏛️ UKAI member · 📜 EU AI Pact signatory
>

>
>

> Zertia is an ANAB-accredited AI management system certification body. The Convention does not produce direct certification engagements — it is a binding treaty addressed to states, not to private actors directly — but it shapes the rights-based floor against which national AI regulation, conformity assessment, and certification work are measured.
>

>
>

> Regulatory frameworks — EU AI Act Conformity Assessment, NIST AI RMF Attestation, ISO/IEC 23894 Risk Assessment, Algorithmic Impact Assessment, Pre-Certification Assessment. For clients operating across multiple Convention Parties — particularly EU, UK, US, and Council of Europe member states — our regulatory frameworks team integrates Convention-aligned vocabulary into compliance positioning. Algorithmic Impact Assessments are structured to satisfy Convention Article 16 obligations on risk and impact management alongside national framework requirements.
>

>
>

> Certification — ISO/IEC 42001, AIUC-1, ISO/IEC 27001, ISO/IEC 27701, ISO/IEC 22301. ISO/IEC 42001 implementation aligns with the Convention’s substantive obligations on transparency, oversight, accountability, reliability, and safe innovation. AIUC-1 operationalises Convention principles for agent-specific deployments where rights-based assurance is material to procurement and deployment decisions.
>

>
>

> Audit — AI Management System audits, High-Risk AI System audits, AI Model audits, EU AI Act audits, NIST AI risk audits. Audit engagements in Convention Parties produce documentation that supports both national framework compliance and Convention-aligned rights protection evidence.
>

>
>

> Training — AI Governance, Data Governance, Privacy Governance through Zertia Academy. Programmes treat the Convention’s rights-based architecture explicitly, with attention to how it interacts with the AI Act in EU territory and with non-EU national frameworks elsewhere. Particularly relevant for organisations whose legal and compliance teams need to position AI governance work across multiple Convention Parties.
>

>
>

> Zertia operates from Boston, Madrid, and London, with ANAB accreditation in the United States and active accreditation processes with UKAS (United Kingdom) and ENAC (Spain/EU). Member of IAPP, INCITS, and UKAI. Signatory to the EU AI Pact.
>

>
>

> ### 🎯 Take action
>

>
>

> | 🔍 Diagnose your gap | 📊 Build the compliance evidence |
>

> |—|—|
>

> | [Pre-Certification Assessment →](https://zertia.ai/regulatory-frameworks/) | [ISO/IEC 42001 Certification →](https://zertia.ai/iso-42001/) |
>

> | Map AI governance against the Convention’s substantive obligations and the international standards (ISO/IEC 42001) that operationalise them. | Accredited certification of the AI management system that operationalises Convention principles into auditable practice across signatory jurisdictions. |
>

>
>

> [Discuss multi-jurisdictional rights-based AI governance positioning →](https://zertia.ai/regulatory-frameworks/)
>

Frequently asked questions

Is the Council of Europe Framework Convention on AI legally binding?

Yes. It is the first international legally binding treaty on AI. Parties undertake obligations of result that they must give effect to through domestic legislative, administrative, or other measures. Failure to satisfy obligations engages international responsibility under treaty law and may produce indirect judicial scrutiny through the European Court of Human Rights for Council of Europe member Parties.

When did the Convention enter into force?

On 1 November 2025, following ratifications by the United Kingdom, France, Norway, and other states meeting the threshold of five ratifications including at least three Council of Europe member states.

Which states have signed and ratified?

Initial signatories at opening on 5 September 2024 included Andorra, Georgia, Iceland, Norway, Moldova, San Marino, the United Kingdom, Israel, the United States, and the European Union. By early 2025, signatures numbered over thirty when counting EU member states represented through the Commission. Ratifications include the United Kingdom, France, Norway, and additional states. The list expands as further Parties complete domestic ratification procedures.

How does the Convention relate to the EU AI Act?

They are complementary. The AI Act is the EU’s mechanism for satisfying Convention obligations within EU territory. Convention obligations are rights-focused; AI Act obligations are product-safety-and-rights-focused. Compliance with the AI Act will, by construction, satisfy most Convention obligations for EU Parties. Non-EU Parties find their own domestic mechanisms for satisfaction.

Does the Convention apply to private actors?

Indirectly. The Convention’s substantive obligations bind public authorities and persons acting on their behalf directly. For private actors, each Party declares at signature how it intends to give effect to the obligations — through direct application of the principles, sectoral regulation, contractual mechanisms, or other appropriate measures. This declaration mechanism is the structural compromise that allowed signatories with different regulatory traditions to commit to the Convention.

What about national security activities?

National security activities are excluded from the Convention’s principal obligations under Article 3(2), provided those activities are conducted consistently with international human rights law and respect democratic institutions and processes. The exclusion was negotiated with significant input from the United States and other parties with substantial intelligence and defence AI capacity.

How is the Convention enforced?

Through the Conference of the Parties as the principal monitoring mechanism, complemented by indirect judicial scrutiny through the European Court of Human Rights for Council of Europe member Parties. The enforcement model follows Convention 108 (data protection) — non-judicial monitoring with rights-based interpretation through national and international courts.

Can non-European states join the Convention?

Yes. The Convention is open to accession by any state in the world. Non-Council of Europe states involved in negotiation included Argentina, Australia, Canada, Costa Rica, the Holy See, Israel, Japan, Mexico, Peru, the United States, and Uruguay. Several signed at opening; others have signed or are considering signature.

Should organisations track the Convention separately from national AI regulation?

Yes, particularly multinational organisations operating across multiple Convention Parties. National AI regulation will increasingly converge on Convention-aligned rights protections, but the Convention itself provides the international floor that national regulation cannot fall below. Tracking the Convention’s Conference of the Parties output and the domestic implementation choices of relevant Parties is part of mature multi-jurisdictional AI governance.

Cross-links

Related glossary terms: [Framework convention](#) · [Council of Europe](#) · [Conference of the Parties](#) · [Convention 108 / 108+](#) · [Obligation of result vs obligation of means](#) · [Declaration mechanism](#) · [Rights-based floor](#) · [European Court of Human Rights](#)

Applicable risk families (AI Risk Repository): [Discrimination and biased outputs](#) · [Privacy violations](#) · [Failure of human oversight](#) · [Lack of transparency and explainability](#) · [Threats to democratic process integrity](#) · [Concentration of power and economic harms](#) · [Harm to rule of law](#)

Related regulations in this Hub: [EU AI Act](#) · [OECD AI Principles](#) · [UNESCO Recommendation on the Ethics of AI](#) · [UK AI Regulation](#) · [Canada AIDA](#) · [US Executive Orders](#) · [GDPR — AI provisions](#)

Zertia services: [Regulatory Frameworks](https://zertia.ai/regulatory-frameworks/) · [Certification](https://zertia.ai/certification/) · [Audit](https://zertia.ai/audit/) · [Training](https://zertia.ai/training/) · [ISO/IEC 42001 Certification](https://zertia.ai/iso-42001/) · [EU AI Act Conformity Assessment](https://zertia.ai/services/eu-ai-act/)

Last reviewed: May 2026 · Sources: Council of Europe Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law (CETS No. 225); Explanatory Report to CETS 225; Council of Europe Treaty Office signature and ratification register; European Parliament Recommendation A10-0007/2026 on EU conclusion of the Convention; US Department of State announcement on signature (5 September 2024); UK ratification announcement; Cambridge International Legal Materials commentary by Marc Rotenberg et al.