Main Insight
Our expert workshop at IASEAI in February 2026 explored how to get to enforceable AI red lines. Three things are required: agreement on thresholds, verification infrastructure, and diplomatic coalition-building. Progress on all three is within reach.
Where Do We Draw the Line? Inside the IASEAI Expert Workshop on AI Red Lines
The AI Red Lines Workshop at IASEAI
AI red lines governance is quickly emerging as a central question in global AI policy. On 26 February 2026, The Future Society and the French Center for AI Safety (CeSIA) convened a four-hour workshop at UNESCO headquarters in Paris, as part of the International Association for Safe & Ethical AI (IASEAI) conference. The event brought together policy practitioners, researchers, technical experts, and civil society representatives to work through one main question: how do we move from broad agreement that some AI capabilities are unacceptable, to verifiable and enforceable international red lines?
Broad Consensus on High-Risk AI Capabilities
The findings were more concrete than many expected. Expert consensus on the most dangerous AI capabilities is broader than the current pace of policy action suggests. A small set of sharp red lines (such as those covering Chemical, Biological, Radiological, and Nuclear (CBRN) uplift, recursive self-improvement without oversight, AI-enabled mass surveillance, and children’s safety) already commands strong agreement among technical and policy communities.
The Primary Constraint on AI Red Lines Governance: Political Will
On verification, workshop participants mostly believe that political will, more than technical feasibility or institutional design, is the greatest obstacle. But this does not mean all states are equally resistant. Middle powers represent an underexplored and urgent source of momentum. Willing states can establish credible governance regimes, build precedent, and make inaction increasingly costly for others, and can encourage global compliance and uptake of equivalent commitments.
The Urgent Need for Verification Infrastructure
Critically, verification infrastructure cannot wait for diplomatic agreement to be reached. The building blocks, including hardware-level attestation, bilateral verification arrangements, and independent audit bodies can take time to construct. The lessons shared by participants from nuclear, biological, and chemical governance are that verification architecture must be built alongside diplomacy, not after it.
Near-Term Opportunities to Advance AI Red Lines Governance
The 2026 diplomatic calendar offers a sequence of concrete opportunities to advance this agenda, including the G7 in France, the Global Dialogue in Geneva, UNGA and the G20 in the USA. According to the workshop’s participants, a minimum viable agreement on concrete AI red lines is achievable and these events present important opportunities.
