AI Liability Insurance 2026: Exclusions, Gaps and New Cover

Quick answer

In 2026, AI risk is moving out of the "silent" coverage of traditional policies and into explicit wording. Verisk's ISO forms introduced optional generative AI exclusions for commercial general liability from January 2026, and many carriers are adding AI exclusions or sub-limits to liability, professional and cyber policies. At the same time, specialist AI products have appeared, covering issues such as AI model underperformance, errors and some regulatory costs. Expect underwriters to ask about your AI inventory, governance, testing and vendor contracts.

As organisations deploy AI in customer-facing and decision-making roles, a practical question follows: if the AI gets it wrong, who pays? This guide explains how insurance responds today, where the gaps are, and how governance affects insurability.

Where AI losses come from

Risk	Example	Traditional policy that might respond
Bodily injury or property damage	AI-controlled machinery or a vehicle causes injury	General or product liability
Professional error	AI-assisted advice or service is wrong and causes financial loss	Professional indemnity / E&O
Discrimination	A biased hiring or lending model	Employment practices liability; D&O
IP infringement	Generative AI output infringes copyright	Media liability; IP cover; sometimes CGL Coverage B
Privacy and security	Model leaks personal data; prompt injection leads to a breach	Cyber
Defamation and misinformation	A chatbot makes false statements about a person	Media liability; CGL Coverage B
Model underperformance	The AI doesn't perform as promised	Often not covered by traditional policies
Regulatory action	Investigation under AI or privacy law	Cyber/D&O (fines often not insurable)

The move to explicit AI exclusions

For years, AI losses were covered, or not, "silently": policies didn't mention AI at all. That's changing.

Verisk/ISO generative AI exclusions (January 2026 edition): optional endorsements for commercial general liability. CG 40 47 excludes Coverage A and Coverage B for harms arising from generative AI. CG 40 48 excludes only Coverage B (personal and advertising injury). CG 35 08 applies to products and completed operations. Insurers using ISO forms can now add these easily.
Carrier-specific wording: some insurers have added AI exclusions or sub-limits to professional, D&O or cyber policies, or require AI-use disclosures at renewal.

Check your renewal wording. An exclusion can be added at renewal with little attention. Ask your broker directly: "Does this policy exclude, limit or specifically cover losses arising from AI, including generative AI we use or provide?"

Specialist AI insurance products

Several insurers and managing general agents now offer AI-specific cover, often called "affirmative" AI cover. Examples include Munich Re's aiSure performance guarantees and products from specialists such as Armilla, often backed by Lloyd's capacity. Typical features:

Performance or warranty cover: pays if an AI model fails to meet agreed performance metrics.
AI errors and omissions: covers third-party claims from AI mistakes, such as hallucinations or incorrect outputs.
Defence and response costs for AI-related claims or regulatory inquiries, within legal limits.

Terms vary widely, and the market is still developing. Read definitions of "AI system", "error" and "performance" carefully.

What underwriters ask

An inventory of the AI systems you develop, deploy or buy.
Your governance: policies, accountable owners, alignment with frameworks such as the NIST AI RMF or ISO/IEC 42001.
Testing and monitoring: bias testing, red teaming, accuracy metrics, drift monitoring.
Human oversight for high-impact decisions.
Vendor contracts: indemnities, warranties, liability caps, IP protections from AI providers.
Incident response plans covering AI failures.
Regulatory exposure, for example EU AI Act high-risk systems or US state AI laws.

Contracts as risk transfer

Insurance is one layer. The other is contracts. Many major generative AI providers now offer IP indemnities for enterprise customers, with conditions attached. Deployers should negotiate warranties, performance commitments, data-use limits and notice of model changes. See AI governance interview questions for how vendor risk comes up in practice.

Which policies might respond to an AI loss

Policy	Possible AI-related cover	Watch for
Commercial general liability (CGL)	Bodily injury or property damage claims	Optional generative AI exclusions introduced by Verisk/ISO in the January 2026 forms
Professional liability / E&O	Claims that AI-assisted services were negligent	Definitions of "professional services"; AI exclusions
Cyber	Data breaches, network security failures, some privacy liability	Whether model failures without a security breach are covered
Directors and officers (D&O)	Claims against directors over AI oversight or "AI-washing" disclosures	Regulatory fine exclusions; conduct exclusions
Employment practices liability	Discrimination claims involving AI hiring tools	Exclusions for algorithmic decision-making
Specialist AI performance cover	Losses when an AI model underperforms against agreed metrics	Strict underwriting conditions and testing requirements

A practical review for risk managers

List your AI uses and the losses each could cause: injury, financial loss, discrimination, privacy, IP, reputation.
Map each loss to existing policies and check definitions and exclusions with your broker.
Ask insurers directly whether AI-related claims are covered, excluded or silent.
Close gaps through contracts (vendor indemnities, limits of liability), controls and, where available, specialist cover.
Prepare evidence of governance (inventory, assessments, testing, monitoring) to support underwriting.

Liability trends to watch

Chatbot cases: courts and tribunals have held companies responsible for their chatbots' statements, as in Moffatt v Air Canada (2024).
Product liability in the EU: the revised Product Liability Directive (2024) covers software, including AI, as a product, with member states applying it from December 2026.
Harm from AI companions and chatbots: lawsuits in the US have tested whether AI output is protected speech and whether developers owe duties to vulnerable users.

Regulation and liability

EU Product Liability Directive (2024 revision): expressly covers software, including AI systems, as products, with strict liability for defects. Member states must apply it from December 2026.
EU AI Liability Directive: the Commission withdrew this proposal in 2025.
EU AI Act fines of up to €35 million or 7% are generally not insurable in many jurisdictions, although defence costs may be.
US: existing tort, consumer-protection and anti-discrimination law, plus state AI laws such as Texas TRAIGA and Colorado's 2027 ADMT law.

AI risk transfer is part of AI governance

The AIGP Complete Pack covers liability, contracts and third-party risk in its Domain II and IV questions, mapped to BoK v2.1.

Get the Complete Pack

Frequently asked questions

Does general liability insurance cover AI?

Sometimes, by default, because older policies don't mention AI. But insurers can now add ISO generative AI exclusions (January 2026 edition), and many are clarifying or excluding AI risk.

What is AI liability insurance?

Specialist cover for losses from AI systems, such as model underperformance, AI errors causing third-party loss, and related defence costs. Products and terms vary.

Can EU AI Act fines be insured?

Generally not, in many jurisdictions, for public-policy reasons. Defence and investigation costs may be insurable.

How can I make my AI more insurable?

Keep an AI inventory, align with recognised frameworks, test and monitor systems, ensure human oversight and negotiate strong vendor contracts.

Does cyber insurance cover AI failures?

It may cover AI-related data breaches and security failures, but often not losses caused by a model simply being wrong without a security incident. Check definitions and exclusions with your broker.

Does the EU Product Liability Directive cover AI?

Yes. The revised Product Liability Directive adopted in 2024 treats software, including AI systems, as a product. Member states must apply the new rules to products placed on the market from December 2026.

Sources

Last reviewed: October 2026. This is general information, not insurance or legal advice.