The Agenda
The full agenda will be as follows:
Day 1
-
08:00 – 09:00
Registration and morning networking
-
09:00 – 09:15
Opening remarks — Welcome to the World AI Regulation Summit
The host and chair open the Summit, frame the proceedings approach, and introduce the rapporteur team.
-
09:15 – 10:15
Opening Keynote and Chair's Dialogue — The Geopolitics of AI Regulation in 2026: Blocs, Standards and the Limits of Alignment
Competing political economies of AI are no longer converging. Standards bodies have become the new diplomatic terrain. Where can jurisdictions still align without surrendering democratic or constitutional differences?
- What kind of authority is sufficient for systems that can learn, generate, recommend, and act autonomously?
- Where can jurisdictions align on governance without giving up democratic or constitutional differences?
- What outcomes should an annual global policy forum deliver that a commercial conference cannot?
Invited / proposed contributors:Host and chair of the Summit, in dialogue with a senior counterpart from a contrasting jurisdiction.
-
10:15- 10:45
Morning coffee
-
10:45 – 11:15
Session 1 — The Global Regulators' Map: What Each Major Jurisdiction Is Building Now
A briefing-format session in which rapporteurs and invited regulators set out where each major jurisdiction currently stands, what is decided, what is still contested, and what will affect industry compliance before the next Summit.
-
11:15 – 11:45
Session 2 — Where the Frameworks Actually Conflict
Rapporteurs identify the genuine friction points: horizontal versus sectoral law, hard versus soft governance, extraterritorial enforcement, auditability and liability. The session surfaces the irreducible disagreements multinationals must now operate around.
11:45 – 12:15Session 3 — Foundation Models and General-Purpose AI: Regulating the Infrastructure Layer
How to govern general-purpose and frontier models without freezing innovation or leaving deployment harms unaddressed. What belongs at the model, deployment and infrastructure layers? How should systemic risk and compute concentration be handled?
12:15 – 12:45Session 4 — Generative AI, Training Data, Copyright and the Global Litigation Wave
Court decisions in the US, UK and EU are setting precedent faster than legislatures can respond. Training data, opt-outs, fair use, text and data mining, output liability and the emerging licensing market are all in play.
12:45 – 13:15Session 5 — Agentic AI and Autonomous Systems: When the System Acts, Who Is Accountable?
AI agents now browse, decide, transact and escalate across workflows. Who owns the decision when an agent acts? What evidence must exist at execution? How should accountability be allocated between developer, deployer, operator and user?
13:15 – 14:15Lunch
Afternoon parallel tracks — Day One
TRACK A — Jurisdictional Experiments TRACK B — Empirical & Field Studies 14:15 – 14:45 EU AI Act high-risk deadline: implementation in practice
Europe's transition from statute to enforcement. GPAI obligations, systemic-risk thresholds, technical documentation and the race to publish standards. What happens when obligation, guidance and capacity mature at different speeds?
Evidence base: AI Office enforcement actions, code-of-practice signatories, technical-standards pipeline, member-state competent-authority designations.
Measuring the deepfake threat: what the data actually says
Quantitative research on synthetic media harm — election interference, financial fraud, image-based abuse — and the gap between media coverage and measured incidence. Regulating on evidence versus regulating on perception.
Evidence base: AI Incident Database, Internet Watch Foundation reports, banking fraud telemetry, election integrity field studies.
14:45 – 15:15 South Korea's AI Basic Act — the first year of a comprehensive law in force
The first major economy outside the EU to operationalise a comprehensive AI law. Transparency duties, domestic representative requirements and the experience of enforcing rules while assurance tooling continues to mature.
Evidence base: KCC and PIPC enforcement records, AI registration data, first-year designation statistics.
South Korea's AI Basic Act — the first year of a comprehensive law in force
Public-sector AI audits in the UK, Netherlands, France and US have produced evidence on bias, error rates, appeal volumes and harm distribution. What the methodologies converge on and what they imply for statutory audit requirements.
Evidence base: ICO algorithmic audits, AlgorithmWatch case files, US federal AI use-case inventories, Dutch ADM findings.
15:15 – 15:45 TRACK A — Jurisdictional Experiments TRACK B — Empirical & Field Studies Singapore AI Verify: pre-deployment assurance in action
A working case study in pre-deployment assurance for general and agentic AI. Testing methodology, governance toolkit, mutual recognition prospects with EU conformity assessment, and running assurance in parallel with statutory development.
Evidence base: AI Verify Foundation testing reports, IMDA Model AI Governance Framework adoption, Project Moonshot generative AI evaluation outcomes.
Bias and fairness audits in employment AI: the empirical picture
Hiring algorithms and workplace monitoring have produced one of the largest bodies of empirical AI research outside healthcare. What bias metrics actually correlate with downstream harm, and what regulators are now asking of vendors.
Evidence base: NYC Local Law 144 bias audit filings, EEOC AI enforcement data, peer-reviewed vendor audits, trade-union submitted evidence.
15:45 – 16:15 Afternoon coffee — joint break 16:15 – 16:45 The UK's principles-based approach: results from the sandbox programmes
The UK's bet on principles-based, sector-led regulation supported by the AI Safety Institute and regulator-led sandboxes. Where principles have held, where they have not, and the case for or against moving to horizontal law.
Evidence base: DRCF AI sandbox outcomes, AISI evaluation findings, ICO and FCA AI enforcement actions, AI Opportunity Action Plan delivery metrics.
Healthcare AI: clinical evaluation, post-market surveillance and patient outcomes
The sector with the longest regulatory track record and deepest evidence base. Diagnostic concordance, off-label use, post-market surveillance failures, and the friction between medical-device regulation and horizontal AI law.
Evidence base: MHRA, FDA and EMA post-market surveillance data; peer-reviewed clinical AI evaluations; NHS AI Lab deployment reviews.
16:45 – 17:15 Japan, Hong Kong and the Gulf: parallel models of strategic AI sovereignty
Japan's agile governance, Hong Kong's data-protection-led framework, and the UAE and Saudi national AI strategies are each producing distinct sovereignty models. What smaller jurisdictions can adapt — and where the models actively conflict.
Evidence base: G7 Hiroshima AI Process returns, Hong Kong PCPD AI guidance compliance, UAE Council for AI metrics, SDAIA national strategy KPIs.
Financial services AI: model risk, fraud detection and supervisory expectations
Banking and insurance supervisors have built model-risk frameworks over two decades. What supervisory examination of AI is finding — explainability gaps, drift, third-party concentration — and how it should inform horizontal AI law.
Evidence base: PRA, FCA, OCC and ECB supervisory reviews on AI models; FSB AI risk assessments; insurance sector AI use-case telemetry.
17:15 – 17:30 Day One synthesis — Rapporteur summary (joint plenary) 19:30Chair's dinner
Invitation-only dinner for invited speakers, advisory contributors and senior delegates. Venue confirmed to invitees.
Day 2
-
08:30 - 09:00
Morning coffee
-
09:00 – 10:00
Opening Keynote — The View from the Lab: What AI Developers Need Regulators to Understand
Senior figures from AI laboratories and infrastructure providers explain — directly to legislators and regulators — where regulation is technically coherent, where it is not, and what meaningful oversight should require.
Invited contributors:Senior keynote from a frontier AI laboratory or AI infrastructure leader, in dialogue with the Summit chair.
-
10:00-10:30
Morning coffee
-
10:30 – 11:00
Session 6 — Policy to Compliance: What Every Industry Must Prepare For
Which obligations will affect all regulated industries: risk assessment, documentation, human oversight, transparency, incident reporting, auditability and vendor governance. What boards should ask of AI vendors, cloud providers and model developers now.
-
11:00 – 11:30
Session 7 — Synthetic Media, Deepfakes, Identity and Digital Trust
Platform takedown obligations, provenance and watermarking standards, identity verification, targeted deepfake offences and cross-border enforcement. The UK, Singapore, US, South Korea and EU compared on what is working and what is theatre.
-
11:30 – 12:00
Session 8 — The United States in 2026: Federal Pre-emption, State AI Laws and the Politics of Fragmentation
A single combined US session. Federal executive direction, the pre-emption debate, the NY RAISE Act, California's transparency obligations and the wave of state AI laws. Is US fragmentation transitional or the new equilibrium?
-
12:00 – 12:30
Session 9 — National Security, Dual-Use Risk, Frontier Controls and Export Regimes
Frontier model evaluation, compute export controls, dual-use risk, deployment to adversarial actors, AI in defence, and the role of AI safety institutes. Where civil AI regulation and national security policy are shaping each other.
-
12:30 – 13:30
Lunch
-
Afternoon parallel tracks — Day Two
Afternoon parallel tracks — Day Two TRACK B — Rights, Redress & Public-Interest Studies 13:30 – 14:00 AI assurance and audit: what the evidence shows the new profession can and cannot do
Documentation, testing, audit trails, monitoring, escalation and board oversight. What regulators will expect to see when AI harms are investigated, and what the assurance industry is actually offering versus what it claims.
Evidence base: Big Four AI audit methodologies, ISO/IEC 42001 certification data, third-party audit pilots, AI assurance market sizing studies.
Rights, redress and public accountability when AI mediates decisions
Discrimination, automated decision-making, public services, welfare, employment and consumer harms. What meaningful redress looks like, whether existing law can cover AI-specific harms, and what audit, explanation and appeal should require.
Evidence base: ECHR AI-related judgments, data protection authority enforcement, ombudsman case files on automated decisions, civil society legal-aid casework.
14:00 – 14:30 Frontier model evaluations: what AI safety institutes have actually measured
UK, US, Japan, Singapore, EU and Canadian AI safety institutes have conducted joint and unilateral evaluations of frontier models. What they have measured — capability uplift, dual-use risk, alignment failures — and what statute can support.
Evidence base: ASI joint testing reports, model system cards and responsible scaling policies, MLCommons safety benchmarks, NIST GenAI evaluations.
AI in the courtroom: judicial pilots and AI in legal practice
Courts in Brazil, Estonia, the UK, China and the US have piloted AI for legal research, case triage and decision support. Accuracy, hallucination, fairness across litigant types, and the emerging consensus on responsible judicial AI use.
Evidence base: Brazilian CNJ Victor and Athos pilot data, Estonian e-Court evaluations, UK Judicial Office AI guidance, US federal court AI working-group findings.
14:30 – 15:00 Incident reporting: building the evidence base for the next decade of regulation
Mandatory AI incident reporting is appearing in the EU AI Act and national frameworks. What aviation, pharmacovigilance and cybersecurity reporting have achieved, and how design choices will determine whether AI incident data is usable.
Evidence base: AI Incident Database, OECD AI Incidents Monitor, aviation and pharmacovigilance reporting evaluations, EU AI Act Article 73 implementation.
AI sovereignty, capacity and the Global South: what works in smaller jurisdictions
What regulatory proportionality, sandboxes, capacity-building and pooled-resource approaches have actually achieved in smaller jurisdictions — and the risk of structural dependency on foreign technology stacks.
Evidence base: African Union AI Continental Strategy data, ASEAN AI Governance Guide adoption, UNCTAD digital readiness assessments, Smart Africa AI outcomes.
15:00 – 15:30 Afternoon coffee — joint break 15:30 – 16:00 Standards diplomacy: ISO, IEEE, ITU, OECD and the politics of interoperability Compute, energy and the political economy of AI infrastructure TRACK A — Assurance, Audit & Compliance Experiments TRACK B — Rights, Redress & Public-Interest Studies Singapore AI Verify: pre-deployment assurance in action
Mutual recognition of conformity assessments, interoperability of risk taxonomies, certification of foundation models, and the role of GPAI and the Council of Europe AI Convention. Whether standards can carry weight hard law cannot.
Evidence base: ISO/IEC JTC 1/SC 42 published standards, IEEE P7000 series adoption, ITU AI for Good outcomes, OECD AI Policy Observatory indicators.
Bias and fairness audits in employment AI: the empirical picture
Compute concentration, data centre siting, energy demand, water use, sovereign cloud and the climate-policy collision course. What regulators should be asking about the physical layer everything else depends on.
Evidence base: IEA AI energy demand projections, national data-centre permitting data, sovereign compute initiative budgets, hyperscaler grid-impact disclosures.
16:00 – 16:30 AI Governance Scenario Exercise — When the Agent Fails
A compressed cross-sector scenario. An autonomous AI system causes simultaneous regulatory, legal, consumer and reputational harm. Pre-assigned roles work the scenario in real time. Output: a one-page AI Incident Governance Checklist.
Evidence base: Scenario derived from anonymised composites of recent enforcement actions and public incident reports.
Civil society audits and independent research: producing evidence legislators can use
Independent audits by civil society, academic centres and investigative journalism have produced some of the most consequential AI evidence of the past five years. What makes such evidence usable in legislative drafting, and its methodological limits.
Evidence base: Ada Lovelace Institute audits, Mozilla and Markup investigative work, Stanford HAI AI Index, academic field experiments cited in legislative drafting.
16:30 – 17:30 Closing Keynote and Proceedings Adoption (joint plenary) The closing session identifies what should be tracked annually and adopts the working direction of the proceedings document. What policymakers and industries need to act on now, who bears responsibility if governance cannot keep pace, and what the next edition should be measured against.
Invited contributors: Closing keynote from a senior AI governance figure of international standing, with response from the Summit chair.
-
17:30 – 17:45
Closing remarks — Next steps and the annual proceedings cycle
Final remarks from the chair, publication timeline for the proceedings document, and invitation to the next annual edition of the Summit.
*This is a working agenda and topics and guests are subject to change.
