AI Indemnification Clauses: What Your Vendor Covers
Covers What major AI vendors' indemnities cover, what the conditions are, and where the defense stops
by
Teddy Ellison
AI & Data Agreements
Summary
AI indemnification is your vendor's promise to defend you and cover the legal bill if a third party sues claiming the model's output infringes their copyright. As of September 2026, OpenAI, Anthropic, Google, and Microsoft all offer one on their paid and business tiers, each covering different products and conditions. It covers the model's output, not your inputs, and outside that you pay the defense and any damages.
What is AI indemnification?
Generative AI indemnification decides who pays when the model's output gets you sued for copyright infringement. With an indemnity, the vendor takes on the defense and the covered costs, not you. That matters because a copyright suit over what your product generated can run past what a seed round covers.
No matter the vendor, the indemnity covers the model's output and stops at your inputs. If a copyright claim comes from a prompt you wrote or data you fine-tuned on, you defend it yourself. It covers infringement claims only. It doesn't cover a product-defect suit, a privacy complaint, or a defamation claim when the model gets a fact wrong.
Indemnity coverage is limited to paid tiers, but today’s AI vendors diverge on what voids it and what it leaves out.
Which AI vendors offer indemnification, and where does each stop?
OpenAI, Anthropic, Google, and Microsoft all offer an output-copyright indemnity to paying business customers, a protection the market often calls a "copyright shield." Each attaches its own conditions, so read your vendor's terms against how your product is built. Everything below reflects the versions in force as of September 2026, with each vendor's own effective date noted, and these terms change often, so verify the live text before you rely on it.
OpenAI indemnification
OpenAI grants the indemnity to its API and ChatGPT Enterprise customers through its Service Terms (updated September 2026), which cover "any third party claim that Customer's use or distribution of Output infringes a third party's intellectual property right." ChatGPT Free and Plus get none from the consumer Terms of Use (effective January 1, 2026), and the only indemnity there runs the other way, making business users defend OpenAI.
OpenAI's indemnity drops away under several conditions. It does not apply if you disabled or ignored OpenAI's citation and filtering features, if you knew or should have known the output was infringing, or if the output "was modified, transformed, or used in combination with products or services not provided by or on behalf of OpenAI."
If your product edits the model's outputs or merges them with your own retrieval layer, you have combined them with services OpenAI didn't provide, and the coverage drops. Check the clause against your architecture before you rely on it. Beta and preview services carry no indemnity at all.
Anthropic indemnification
Anthropic's Commercial Terms (effective June 2025) indemnify paid use only, defined as "Customer's paid use of the Services," and cover not just the output but claims over "data Anthropic has used to train a model." The coverage carries no dollar cap. The terms limit overall liability to the previous twelve months of fees, then exempt the indemnity from that ceiling, so you keep the defense even when a claim dwarfs your Anthropic bill.
Its exclusions track OpenAI's. Modified or combined outputs, your own inputs, and trademark use in commerce are all carved out, plus one more for claims that an output practices a patented invention. The same section makes indemnification "each party's sole and exclusive remedy," so if an exclusion applies, you have no fallback claim against Anthropic.
Google generative AI indemnification
Google ties its coverage to a published list. The Generative AI Indemnified Services page (last revised July 2026) names what's covered, including the Gemini Enterprise Agent Platform API on generally available versions of Gemini, Imagen, and Veo, plus Gemini Enterprise, NotebookLM Enterprise, and Gemini in Workspace. A preview model or an off-list service gets no output indemnity, so recheck the list when your stack changes.
Microsoft AI indemnification
Microsoft's Customer Copyright Commitment (last updated March 2026) covers output from Copilot products and Azure OpenAI under its Product Terms. On Azure OpenAI, coverage depends on Microsoft's required mitigations. These include a system prompt steering the model away from infringing output and a retained testing report proving you checked your app for third-party reproduction. Microsoft says a customer tendering a claim "will be required to demonstrate compliance," so build that report in advance.
Final Thoughts
Most of this you can handle yourself. On a paid or business tier with an unmodified pipeline, you're covered for the case that matters most, a third-party claim that the model's raw output infringes a copyright. Reading your vendor's terms to confirm it is well within reach.
A second set of eyes earns its keep at the seam between contracts. When you warrant IP to an enterprise customer, or an acquirer's counsel asks exactly what your vendor covers, the distance between the vendor's carve-outs and your own promises becomes your personal exposure. Getting that read before the deal beats reconstructing it under one. Our guide to who owns AI-generated content covers the carve-outs and the downstream flow.
Reach out to set up a free consultation.
Serotonin Legal advises technology founders on corporate, regulatory, and transactional matters at the intersection of AI, blockchain, and fintech. This guide is for general informational purposes and does not constitute legal advice. No attorney-client relationship is formed by reading this material.
FAQs
What is an AI indemnification clause?
An AI indemnification clause is the contract provision that moves the defense of a third-party IP claim onto the AI vendor. When it applies, the vendor runs the defense and pays the covered costs and damages tied to the output's alleged infringement. Read its conditions and carve-outs before you count on the protection for your use.
Does AI indemnification cover my own inputs or training data?
No. These indemnities carve out the material you supply. OpenAI excludes Output from Input or fine-tuning files the customer "did not have the right to use," and Anthropic excludes "Inputs or other data provided by Customer," and you represent that you hold the rights to whatever you submit. If the infringement comes from your data rather than the model's output, you defend it.
Do free AI tools come with indemnification?
Generally no. Vendor IP indemnities are a paid and business-tier feature. OpenAI's indemnity covers its API and ChatGPT Enterprise customers, ChatGPT Free and Plus get none, and Anthropic's coverage is defined by paid use. If your team ships production work through a consumer login, the output carries no indemnity, so move that traffic to the API or an enterprise agreement before you rely on the coverage.
Do image tools and enterprise AI platforms offer indemnification?
Yes, and the same logic applies. Adobe Firefly, Shutterstock, and Canva indemnify AI-generated images on their enterprise plans. IBM watsonx, AWS Bedrock, and Databricks cover output from their listed foundation models. Each applies to paid or enterprise use, covers third-party IP claims over the output, and carves out modified output, your own inputs, and disabled filters. Check the specific tool's terms first.
When does AI indemnification not apply?
Several conditions can void AI indemnification, and they vary by vendor. The common ones are output you modified or combined with non-vendor tools, your own inputs, disabled filters where the vendor requires them, consumer-tier use, and beta or preview features. OpenAI also drops coverage for output you knew was infringing. Check your vendor's exact conditions before you rely on the indemnity.
Curious to learn more about Serotonin Legal? —
Get in Touch
AI Indemnification Clauses: What Your Vendor Covers
Covers What major AI vendors' indemnities cover, what the conditions are, and where the defense stops
by
Teddy Ellison
AI & Data Agreements
Summary
AI indemnification is your vendor's promise to defend you and cover the legal bill if a third party sues claiming the model's output infringes their copyright. As of September 2026, OpenAI, Anthropic, Google, and Microsoft all offer one on their paid and business tiers, each covering different products and conditions. It covers the model's output, not your inputs, and outside that you pay the defense and any damages.
What is AI indemnification?
Generative AI indemnification decides who pays when the model's output gets you sued for copyright infringement. With an indemnity, the vendor takes on the defense and the covered costs, not you. That matters because a copyright suit over what your product generated can run past what a seed round covers.
No matter the vendor, the indemnity covers the model's output and stops at your inputs. If a copyright claim comes from a prompt you wrote or data you fine-tuned on, you defend it yourself. It covers infringement claims only. It doesn't cover a product-defect suit, a privacy complaint, or a defamation claim when the model gets a fact wrong.
Indemnity coverage is limited to paid tiers, but today’s AI vendors diverge on what voids it and what it leaves out.
Which AI vendors offer indemnification, and where does each stop?
OpenAI, Anthropic, Google, and Microsoft all offer an output-copyright indemnity to paying business customers, a protection the market often calls a "copyright shield." Each attaches its own conditions, so read your vendor's terms against how your product is built. Everything below reflects the versions in force as of September 2026, with each vendor's own effective date noted, and these terms change often, so verify the live text before you rely on it.
OpenAI indemnification
OpenAI grants the indemnity to its API and ChatGPT Enterprise customers through its Service Terms (updated September 2026), which cover "any third party claim that Customer's use or distribution of Output infringes a third party's intellectual property right." ChatGPT Free and Plus get none from the consumer Terms of Use (effective January 1, 2026), and the only indemnity there runs the other way, making business users defend OpenAI.
OpenAI's indemnity drops away under several conditions. It does not apply if you disabled or ignored OpenAI's citation and filtering features, if you knew or should have known the output was infringing, or if the output "was modified, transformed, or used in combination with products or services not provided by or on behalf of OpenAI."
If your product edits the model's outputs or merges them with your own retrieval layer, you have combined them with services OpenAI didn't provide, and the coverage drops. Check the clause against your architecture before you rely on it. Beta and preview services carry no indemnity at all.
Anthropic indemnification
Anthropic's Commercial Terms (effective June 2025) indemnify paid use only, defined as "Customer's paid use of the Services," and cover not just the output but claims over "data Anthropic has used to train a model." The coverage carries no dollar cap. The terms limit overall liability to the previous twelve months of fees, then exempt the indemnity from that ceiling, so you keep the defense even when a claim dwarfs your Anthropic bill.
Its exclusions track OpenAI's. Modified or combined outputs, your own inputs, and trademark use in commerce are all carved out, plus one more for claims that an output practices a patented invention. The same section makes indemnification "each party's sole and exclusive remedy," so if an exclusion applies, you have no fallback claim against Anthropic.
Google generative AI indemnification
Google ties its coverage to a published list. The Generative AI Indemnified Services page (last revised July 2026) names what's covered, including the Gemini Enterprise Agent Platform API on generally available versions of Gemini, Imagen, and Veo, plus Gemini Enterprise, NotebookLM Enterprise, and Gemini in Workspace. A preview model or an off-list service gets no output indemnity, so recheck the list when your stack changes.
Microsoft AI indemnification
Microsoft's Customer Copyright Commitment (last updated March 2026) covers output from Copilot products and Azure OpenAI under its Product Terms. On Azure OpenAI, coverage depends on Microsoft's required mitigations. These include a system prompt steering the model away from infringing output and a retained testing report proving you checked your app for third-party reproduction. Microsoft says a customer tendering a claim "will be required to demonstrate compliance," so build that report in advance.
Final Thoughts
Most of this you can handle yourself. On a paid or business tier with an unmodified pipeline, you're covered for the case that matters most, a third-party claim that the model's raw output infringes a copyright. Reading your vendor's terms to confirm it is well within reach.
A second set of eyes earns its keep at the seam between contracts. When you warrant IP to an enterprise customer, or an acquirer's counsel asks exactly what your vendor covers, the distance between the vendor's carve-outs and your own promises becomes your personal exposure. Getting that read before the deal beats reconstructing it under one. Our guide to who owns AI-generated content covers the carve-outs and the downstream flow.
Reach out to set up a free consultation.
Serotonin Legal advises technology founders on corporate, regulatory, and transactional matters at the intersection of AI, blockchain, and fintech. This guide is for general informational purposes and does not constitute legal advice. No attorney-client relationship is formed by reading this material.
FAQs
What is an AI indemnification clause?
An AI indemnification clause is the contract provision that moves the defense of a third-party IP claim onto the AI vendor. When it applies, the vendor runs the defense and pays the covered costs and damages tied to the output's alleged infringement. Read its conditions and carve-outs before you count on the protection for your use.
Does AI indemnification cover my own inputs or training data?
No. These indemnities carve out the material you supply. OpenAI excludes Output from Input or fine-tuning files the customer "did not have the right to use," and Anthropic excludes "Inputs or other data provided by Customer," and you represent that you hold the rights to whatever you submit. If the infringement comes from your data rather than the model's output, you defend it.
Do free AI tools come with indemnification?
Generally no. Vendor IP indemnities are a paid and business-tier feature. OpenAI's indemnity covers its API and ChatGPT Enterprise customers, ChatGPT Free and Plus get none, and Anthropic's coverage is defined by paid use. If your team ships production work through a consumer login, the output carries no indemnity, so move that traffic to the API or an enterprise agreement before you rely on the coverage.
Do image tools and enterprise AI platforms offer indemnification?
Yes, and the same logic applies. Adobe Firefly, Shutterstock, and Canva indemnify AI-generated images on their enterprise plans. IBM watsonx, AWS Bedrock, and Databricks cover output from their listed foundation models. Each applies to paid or enterprise use, covers third-party IP claims over the output, and carves out modified output, your own inputs, and disabled filters. Check the specific tool's terms first.
When does AI indemnification not apply?
Several conditions can void AI indemnification, and they vary by vendor. The common ones are output you modified or combined with non-vendor tools, your own inputs, disabled filters where the vendor requires them, consumer-tier use, and beta or preview features. OpenAI also drops coverage for output you knew was infringing. Check your vendor's exact conditions before you rely on the indemnity.
Curious to learn more about Serotonin Legal?
Get in Touch
