Benefits and Risks: What Armenia's Accession to the Council of Europe's AI Convention Means for Builders
On January 27 in Strasbourg, Armenian Foreign Minister Ararat Mirzoyan and Council of Europe Secretary General Alain Berset signed the Council of Europe Framework Convention on Artificial Intelligence, Human Rights, Democracy and the Rule of Law.
This is the first international legal instrument focused on protecting fundamental rights in AI use. For engineering leaders, founders, and researchers, it sets the context you'll be shipping into over the next few years.
What the Convention Actually Regulates
Tigran Jrbashyan, Director of Management Consulting at Ameria, makes a key point: the Convention does not cap compute, restrict GPU use, or dictate model sizes. It's not a technical throttle.
Instead, it creates a legitimacy layer-"an important foundation of international trust"-showing Armenia's AI work aligns with human rights and the rule of law. That matters to EU universities, research consortia, and investors who care about ethical AI.
Upside for Armenia's Tech Ecosystem
- Easier collaboration with Europe: More doors open for grants, joint labs, and data-sharing under clear rights and safeguards.
- Investor confidence: Governance signals reduce perceived risk for funds and corporates with compliance mandates.
- Talent magnet: Researchers prefer ecosystems where experiments can run inside trusted, rights-aware frameworks.
The Real Risk: Overreach in Local Policy
The danger isn't the Convention-it's how it could be applied domestically. Jrbashyan warns that if it's treated as heavy-handed technical regulation, it could chill experimentation and slow startups through legal ambiguity.
The strength of the Convention is that it's a high-level guide. It encourages controlled, sandbox-style trials and regulates specific use cases, not the base technology or infrastructure.
What This Means for Teams Building AI in Armenia
- Design for accountability: Keep concise audit trails of data sources, training decisions, evaluation methods, and human-in-the-loop controls.
- Risk and impact checks: Add lightweight impact assessments for higher-risk deployments (health, finance, public sector). Document mitigations.
- Guardrails by use case: Tune controls to application risk level rather than blanket rules on models or compute.
- Join sandboxes: Prefer pilots in regulatory sandboxes to de-risk approvals and clarify expectations early.
- User-facing transparency: Provide clear model cards, limitations, and escalation paths where outcomes affect rights.
- Data governance: Track provenance, consent, deletion, and access. Automate where possible.
- Contracts that travel: Use clauses compatible with EU partners (data, monitoring, incident response, audit access).
Suggestions for Policymakers and Institutions
- Keep it principle-based: Avoid rules that target compute, GPUs, or generic model sizes.
- Protect research: Carve-outs or safe harbors for non-commercial and early-stage research with proper oversight.
- Scale by risk: Clear tiers for low-, medium-, and high-risk use cases with proportionate requirements.
- Make sandboxes the default: Fast-track programs for startups and academia with transparent criteria.
- One-stop guidance: A single portal for templates, checklists, and rulings to cut legal uncertainty.
- Feedback loops: Sunset clauses and iteration based on real-world evidence from pilots.
Why This Matters Now
Handled well, the Convention becomes a competitive advantage for Armenia. Misapplied, it turns into self-limitation. As Jrbashyan puts it, "Everything depends on how we translate it into policy and institutional practice."
Further reading
- Council of Europe: Framework Convention on AI
- Skill up your team for responsible AI development (courses by job)
Your membership also unlocks:
