There is no single global regime for autonomous AI agents. The European Union has codified operator duty under the AI Act. The Colorado AI Act activates before it. The United Kingdom regulates sectorally through existing authorities. Singapore, Japan, Korea, Canada, and Brazil are writing their own texts. An organisation deploying an agent across borders in 2026 is not choosing between regimes. It is accruing obligations under all of them. This publication tracks each one, in its own terms, and in relation to the others.
The fragmentation is not only geographic. It is temporal. Each jurisdiction activates on its own schedule, and the duties accrue even when enforcement is delayed. The below is not exhaustive. It is the subset that binds cross-border agent deployers first.
First comprehensive US state AI statute creating a duty of care for developers and deployers of high-risk AI systems.
European supervisor publishes first positions on underwriting agentic AI. Read globally as reference for insurer exposure.
Article 26 binds any organisation deploying a high-risk AI system inside the Union. Extraterritorial reach if outputs are used in the EU.
Enhanced testing toolkit and Model AI Governance Framework revisions for generative and agentic systems.
Record keeping, human oversight, and incident reporting must be operational and auditable.
Long-form analysis of the statutes, frameworks, and standards shaping how the world will hold operators of autonomous AI agents accountable.
A reading of SB 24-205 as it enters force on 1 February 2026. The duty of care, the algorithmic discrimination standard, the risk management programme, and the comparison with the EU AI Act.
Why a voluntary federal framework is becoming the benchmark courts, regulators, and contracts use to define whether an operator acted reasonably when deploying an autonomous agent.
A structured comparison of how the three largest regulatory spheres are allocating responsibility across the AI supply chain. Horizontal statute, revised products regime, and sectoral regulator leadership.
Singapore's AI Verify, Japan's AI Promotion Act, Korea's AI Basic Act, and why the region's softer-law approach is setting de facto standards many US and EU deployers will follow.
Govern, map, measure, and manage. The four functions are a distillation of what it means to act reasonably when deploying AI that affects people and institutions.NIST Artificial Intelligence Risk Management Framework 1.0 · National Institute of Standards and Technology
Three structural features organise our coverage. First, AI agent liability is accruing before any single regime is mature. Operators deploying systems today face duties that will only be litigated later. The task is not to choose a legal standard. It is to reach the highest one that applies.
Second, the texts converge more than they diverge. Governance, risk management, transparency to affected parties, human oversight, logging, and incident response appear in almost every serious draft. A deployer that builds to the EU AI Act will be largely compliant with the Colorado AI Act, the Singapore Model Governance Framework, and the expectations of UK sectoral regulators. Frameworks differ at the edges, not the centre.
Third, the insurance market is following the regulation, not leading it. Specialty AI coverage exists. It is fragmented, expensive, and underwritten against the same operator duties the statutes describe. The rational posture is to harden the deployment and insure the residual.
The table below is a reduced view of ten jurisdictions and the instrument that carries the primary obligation for AI agent deployers. The column "standard" indicates whether the regime imposes a rule (statutory duty), a framework (voluntary but influential), or a sectoral approach (existing law applied by existing regulators).
| Jurisdiction | Primary Instrument | Standard | Activation |
|---|---|---|---|
| European Union | Regulation (EU) 2024/1689 · AI Act | Horizontal rule | 2 August 2026 |
| United States · Colorado | SB 24-205 · Colorado AI Act | Statute | 1 February 2026 |
| United States · Federal | NIST AI RMF 1.0 · EO on AI | Framework | Continuous |
| United Kingdom | White Paper 2023 · Regulator guidance | Sectoral | Rolling |
| Singapore | Model AI Governance Framework · AI Verify | Framework | Continuous |
| Japan | AI Promotion Act 2024 · METI guidelines | Framework | 2025 rolling |
| Korea | AI Basic Act 2024 | Statute | January 2026 |
| Canada | AIDA (Bill C-27, reframed 2025) | Statute pending | Expected 2026-2027 |
| Brazil | PL 2338/2023 · Marco Legal da IA | Statute pending | Under deliberation |
| China | Interim Measures for Generative AI · CAC rules | Regulation | In force |
Read the full country-by-country review on the jurisdictions page. Each entry links to the primary instrument and the most recent guidance.
Agent Liability sits inside a network of five sister publications covering the regulatory, certification, and insurance dimensions of autonomous AI agent deployment. This is the global desk. Agent Liability EU carries the European regulatory analysis.
A published framework from Future Proof Intelligence for assessing autonomous AI agent deployments. Seven dimensions. Independent. Continuously maintained.
Read the framework