A formal registry for organisations preparing to insure autonomous AI systems ahead of the Union's enforcement deadlines. Coverage opens in alignment with the Artificial Intelligence Act and the revised Product Liability Directive.
The EU AI Act high-risk obligations may shift to 2 December 2027 if the Digital Omnibus is adopted. The insurance market does not pause. Capacity, exclusions, and product launches continue on their own clock. The PLD transposition deadline of 9 December 2026 is unchanged.
AI Act adopted. Operator provisions enforceable on 02.08.2026. PLD 2024 follows on 09.12.2026.
AIUC-1 precedent only. Reinsurer programmes active. Primary market pending.
ISO 42001, NIST AI RMF, AIUC-1, Agent Certified methodology publicly available.
Most European operators have not yet audited their agent stack against Article 26 obligations.
Figure A. The Agent Insured Market Readiness Index. Composite of regulatory activation, carrier activity, standards maturity, and operator preparation. Updated weekly as the European market approaches the August 2026 deadline. Methodology available on request.
Every specialist carrier writing meaningful AI liability limits in 2026, in one interactive reference. Filter by segment, geography, and product posture. Updated quarterly. Inclusion is editorial only, no carrier pays for placement.
From August 2026, the European Union treats the deployment of autonomous AI systems as a regulated activity. From December 2026, the harms they produce sit inside product liability law. Together these two instruments turn AI agents into an insurable class of risk, and expose every operator that has not secured coverage.
Providers and deployers of general-purpose AI systems, and operators of high-risk AI under Annex III, face conformity, transparency, and post-market monitoring obligations. Administrative fines reach EUR 35 million or 7% of global turnover.
Software, including AI systems and the data they rely on, is formally treated as a product. Claimants gain disclosure rights and a rebuttable presumption of defectiveness where an AI system is shown to have contributed to damage.
Underwriting is anchored on certification evidence and deployment telemetry. Organisations on the pre-launch registry are invited into binding quotation in the order of their registration.
An AI agent policy is not a single cover. It is a composite of three risk pillars that together meet the evidentiary expectations of Article 26 of the AI Act, Article 10 of the revised Product Liability Directive, and the four underwriting questions every European AI insurer has started to ask. The seven Agent Certified dimensions feed the pillars. The pillars compose the policy.
Underwriter scrutiny on model reliability, evaluation coverage, failure modes and provenance of data the agent reasons over. The most scrutinised pillar at submission.
Board-level accountability, oversight staffing, role definition, and the autonomy envelope the operator has set for the agent. The pillar regulators focus on.
Who invokes the agent, under what authority, with what approval and rollback controls. The pillar that defines first-notice-of-loss triggers.
No policies are being sold today. This is the indicative framework insurers have begun to price against. An Agent Certified tier discount reduces the loading that autonomy and sector exposure introduce, which is the single largest mechanism operators can use to move the number down before the Q3 2026 coverage window.
An AI system that can act on its own behalf will eventually produce a loss that nobody expected. The work of the coverage market is to decide, in advance, who carries that loss.
Organisations on the registry are notified in advance of underwriting guidance, receive the weekly Agentic Liability Monitor, and are invited into binding quotation before general availability.
Long-form notes on how AI agent liability is being priced, what the emerging insurance standards actually cover, and how European enterprises should prepare for the August and December 2026 deadlines.
When AI agents execute transactions, send contracts, or book services without human review, existing cyber and E&O policies frequently do not cover the resulting losses. A structured analysis of the three failure modes, what Munich Re, Armilla, and AIUC-1 each cover, and how to build a coverage-eligible agentic deployment.
ElevenLabs became the first AIUC-1 policyholder in February 2026 after 5,000+ adversarial simulations. Three structural gaps make the US standard problematic for European operators: Article 99 penalties are not insurable, the Product Liability Directive introduces strict liability, and AIUC-1 does not produce the Annex IV documentation the EU AI Act requires.
Most organisations deploying AI agents have PI and cyber insurance and assume coverage. The assumption is often wrong. This guide maps the coverage triggers, the gap between existing policies, and the decision framework for each AI deployment type.
Counterpart's November 2025 endorsement names AI hallucinations, misclassification, and hiring bias as explicit covered perils. This guide explains the coverage structure, the governance documentation required to qualify, and the European access alternatives for enterprises that cannot reach Counterpart directly.
Standard tech E&O is retreating from AI risk through sublimits and exclusions. Specialist carriers including Armilla, Vouch, Embroker, and Counterpart are building affirmative AI E&O in its place. A guide to who covers what, how to read your policy, and when standalone coverage is necessary.
Twelve specialist and specialist-adjacent carriers now write meaningful AI liability limits. This is the complete reference: every carrier profiled, every capacity figure sourced, every underwriting criterion documented alongside the generalist carriers filing exclusions.
The AI liability market is in active price discovery. This analysis maps rate benchmarks by deployer tier, names the carriers writing meaningful limits, and identifies the five underwriting factors that move premium the most.
The five components of a complete underwriting submission, the timeline for organisations targeting Q3 2026 coverage, and the gaps most operators discover too late to fix before August.
How Armilla's position as a Lloyd's coverholder enables structured AI liability underwriting, and what the coverholder model means for European operators seeking coverage before the 2026 deadlines.
A close reading of the five categories of loss that the first European AI agent liability policies will address, with references to AIUC-1, Munich Re aiSure, and Armilla.
Existing cyber and professional indemnity wordings are being rewritten to exclude autonomous AI activity. A guide to the new exclusion language and how to read a renewal.
A practical preparation note for compliance, risk, and legal teams: what to document, what to ask of insurers, and how certification feeds underwriting.
Munich Re has written AI performance coverage since 2018. aiSure settles on measurable performance data rather than loss adjustment. A guide to what it covers and where the gaps remain.
Agent Liability EU is the operator desk on those instruments. Read the obligations that put this coverage on the European agenda.
agentliability.eu → The methodologyAgent Certified is the published methodology that feeds the Coverage Architecture above. Review the scoring rubric before submitting a registration.
agentcertified.eu →