AI Indemnification

AI indemnification is a contractual promise from an AI vendor to defend and pay for certain third-party claims that arise from a customer's use of the vendor's AI service. The most visible examples are intellectual property indemnifications from foundation model and cloud providers: Microsoft's Copilot Copyright Commitment, Google Cloud's Generative AI Indemnification, Adobe Firefly's enterprise indemnity, and similar commitments from OpenAI for enterprise customers and from Anthropic for commercial Claude customers (API and Claude for Work or Enterprise).

The structure is contract law, not insurance, and the distinction matters. An indemnification is a promise from one private party to another that the indemnifying party will defend and pay; it depends on the indemnifying party's solvency, the contract's exclusions and caps, and the indemnified party's compliance with the conditions (using the service as licensed, not turning off filters or guardrails, not engaging in conduct excluded by the agreement). Insurance, by contrast, is regulated capital backed by a financially strong carrier and is portable across vendors.

AI indemnifications typically cover a narrow slice of the deployer's total AI exposure. The covered claims are usually limited to third-party intellectual property infringement (copyright, sometimes patent or trademark) arising from output the vendor's model generated. They generally do not cover hallucinations, defamation, unauthorized data disclosure, discrimination from algorithmic decisions, regulatory action, or harm caused by an AI agent's autonomous actions. They also condition coverage on the customer using the service within the vendor's terms (which often means keeping safety filters enabled and not feeding the model copyrighted training data).

For an enterprise deploying generative AI across multiple vendors, indemnifications are useful but partial. They reduce IP risk on the specific vendor's service while leaving the broader exposure (across hallucinations, agentic actions, multi-vendor stacks, regulatory action) uncovered. Generative AI Liability insurance is the structural answer to the residual exposure: it sits on top of vendor indemnifications and responds where contract law does not.