AI Liability Insurance

Third-party cover for the liability a company takes on when the AI it deploys causes harm to someone else.

AI liability insurance is the market's name for third-party cover that responds when the AI a company has deployed causes harm to someone outside the company. In practice today it means generative AI liability insurance, because generative AI is the exposure carriers are actually writing against, and the two terms describe the same coverage.

It answers a specific question: who pays when an AI system a business put into production produces an output that injures a third party. The harm can be a financial loss caused by a wrong or fabricated answer, intellectual property infringement or defamation in generated content, the unauthorized disclosure of protected information through an output, or in narrow cases bodily injury or property damage following from what the system produced.

The buyer is the deployer, not the model developer. A company that puts an AI assistant in front of customers, or uses generated content in its products, carries the liability for what that system tells people. This is what separates the cover from Tech E&O, which is written for the vendor that sells software under contract, and from Cyber, which is written for attacks on systems and breach response.

The gap exists because the forms a deployer already carries were not written for it. Verisk and ISO have filed general liability exclusions (CG 40 47, which removes both Coverage A and Coverage B, and CG 40 48, which removes Coverage B only) with a January 2026 edition date, alongside a products and completed operations form (CG 35 08). Where a carrier attaches one, the underlying Commercial General Liability policy no longer answers generative AI claims at all.

Policies in this category are typically written as standalone, claims-made third-party liability forms that schedule the insured's generative AI systems and coordinate with existing cyber and E&O through an Other Insurance clause. Testudo writes one such policy for U.S. companies deploying generative AI, on Lloyd's of London paper, distributed on an excess and surplus lines basis in the United States, subject to underwriting and not available in all states.

Also known as

Generative AI Liability Insurance, Generative AI Third-Party Liability Insurance, Gen AI Liability Insurance, AI Liability Cover

Frequently asked

Is AI liability insurance the same as generative AI liability insurance?

Yes. They name the same coverage. "AI liability insurance" is the category term the market has settled on, and "generative AI liability insurance" is the more precise name for the form carriers actually write, because the policy schedules the generative AI systems it covers. A policy sold under either name responds to third-party claims arising from what the insured's deployed AI produced.

What does AI liability insurance not cover?

Standalone forms respond to third-party harm caused by the generative AI systems scheduled on the policy. They do not act as cyber insurance: losses caused by a cyber-attack sit with the cyber insurer. They are not written for a company's internal-only AI use, where no third party receives or relies on an output. And they respond to the systems on the schedule, so AI a company deploys without scheduling it may not be covered.

Who buys AI liability insurance?

U.S. companies, from mid-market to large enterprise, that have deployed generative AI where it interfaces with a third party or produces output a third party receives or relies on. Customer-facing assistants, generated content in a product, and AI-assisted advice are the common triggers. It is placed through insurance brokers rather than sold direct.

Which insuring agreements appear on an AI liability policy?

A typical standalone form carries six named insuring agreements: Generative AI Errors (financial loss from a wrong or hallucinated output), Intellectual Property Infringement and Personal Injury (copyright, defamation, advertising injury in AI-generated content), Unauthorized Data Disclosure (PII or trade secrets leaked through an output), Bodily Injury (physical harm from reliance on a generative AI output), Property Damage (damage to third-party property caused by action taken on a generative AI output), and AI Regulatory Proceedings (defense costs, plus civil fines and penalties where insurable by law, when a regulator brings an action alleging a violation of an AI regulation). Each is sub-limited and coordinates with the insured's underlying cyber and E&O through an Other Insurance clause.

Who in the AI stack actually buys the cover?

The deployer buys it. That is the enterprise that put the AI system into production for its own customers or operations, whether the underlying model is OpenAI, Anthropic, Google, or self-hosted. Foundation model developers carry separate D&O, professional liability, and IP insurance; the deployer's exposure is to its own users and depends on how the AI is used in production, which the foundation model developer's policies do not address.

Is AI liability insurance claims-made or occurrence?

Almost universally claims-made. The exposure profile (long-tail IP and defamation suits, regulatory action months after deployment, hallucination harms that surface in litigation years later) maps to claims-made structures the way other professional liability lines do. Retroactive dates anchor when covered AI deployments began, and extended reporting periods (tail coverage) are commonly negotiated at non-renewal to preserve cover for incidents already in the pipeline.

Related terms

General information, not legal or insurance advice.