
Which AI tools indemnify you, and what can you do when yours does not?
Several large AI platforms indemnify paying business customers against copyright claims about their output, among them OpenAI, Microsoft, Google, Anthropic, Amazon, Adobe, IBM, Getty Images and Shutterstock, usually excluding free plans, modified output, your own inputs and trademarks. Most AI video and voice tools offer none. You do not need an enterprise plan to be protected. You can buy media liability insurance that affirmatively covers AI content and offer your client your own capped indemnity, which buyers commonly accept in place of the tool’s.
Tell us the situation.
One licensed human replies the same business day — not five agents, not an auto-responder. If the policy you have is already the right one, we will tell you.
We use this to quote and service your insurance, and we do not sell it or pass it to lead networks. Privacy policy.

Kevin Kelsey
Founder | Principal Agent
Edited television for fifteen years and was Emmy-nominated for The Muppets Mayhem. He met network delivery rules from the production side. He is also a software developer. He owns two post-production software products, Baton and B/side. So your AI tools list and your review process are read by someone who builds software and has delivered shows.
Send us what you have. We’ll tell you if it’s right.
Your current policy paperwork, a renewal notice, or nothing at all. A licensed agent reads it and replies the same business day.
What is an AI indemnity?
An AI indemnity is a promise in an AI tool’s terms that the vendor will defend you, and pay a judgment or settlement, if someone claims the tool’s output infringes their copyright or other intellectual property. Microsoft calls its version the Customer Copyright Commitment, OpenAI calls its version Copyright Shield, and Google, Anthropic, Amazon, Adobe, IBM, Getty Images and Shutterstock carry their own.
Read who the promise runs to. It protects the vendor’s customer, the person or company holding the account. If you are a studio or post vendor using the tool for a client, the indemnity protects you, not your client. Your client gets protection only through what you promise them in your own contract.
Which plans carry the indemnity?
Paid business plans, almost without exception. Microsoft excludes free and consumer products. Google requires paid use and excludes credits and free tiers. Anthropic covers paid commercial use, not its consumer plans. Amazon excludes its free developer tier. Adobe sells the indemnity as an enterprise entitlement, and Shutterstock and Canva limit theirs to enterprise customers.
Most AI video, image and voice tools built for creators offer no indemnity at all in their standard terms. Midjourney, Stability AI, Runway, Luma Dream Machine, Pika, Kling, ElevenLabs, HeyGen and Moonvalley each ask the user to indemnify them instead. Some sell enterprise agreements whose terms are not published, so if you are on one, read your own contract rather than the public terms.
What conditions come with an AI indemnity?
The same few appear in nearly every one. The vendor’s safety systems, content filters and citation features must be left on. You must not have known, or had reason to know, that the output infringed. You must have had the rights to whatever you put in. Trademark claims about using the output in commerce are commonly excluded. Notice must be prompt, and the vendor usually controls the defense.
Two conditions matter most for production work. Many indemnities cover only unmodified output, and Adobe says expressly that edits made in Creative Cloud fall outside. Several exclude your own inputs or anything combined with other material. Post, VFX and animation work is almost entirely modification and combination, frequently on footage the client supplied, so an enterprise indemnity often does not reach the work a vendor actually delivers.
Do you have to buy an enterprise plan to be protected?
No. An enterprise plan is one way to get a vendor’s indemnity, and it is not the only way to be protected. You can buy the protection yourself, as a media liability policy that affirmatively covers AI-generated content. That policy defends and pays covered claims that your content infringed a copyright or trademark, defamed someone, or misused a name, likeness or voice, whichever tool made it and whether or not that tool indemnifies anyone.
In some ways it is the broader protection. A media liability policy can respond to edited and combined work, to defamation and privacy claims that tool indemnities do not address, and to work made with several tools at once. A tool indemnity covers one vendor’s output, on that vendor’s conditions. The policy wording governs what is covered, so read it for an AI exclusion, a music exclusion and the start date for past work before you rely on it.
What does your own insurance do, and what does it not do?
It covers your liability. That includes liability you take on in a contract, such as an indemnity to your client, to the extent the policy wording covers it. Many media and professional forms limit liability assumed under contract to what you would have owed anyway, or to specific kinds of claims, so ask whether your form covers an indemnity for infringement claims about your content.
It does not make the AI vendor indemnify anyone, and it does not turn a tool’s consumer terms into enterprise terms. It does not satisfy a clause about where the client’s material goes or whether the vendor trains on it. It also assumes you cleared what you used, so it does not replace music, stock or likeness clearances.
What do buyer contracts ask for, and what do they accept instead?
Buyer contracts for AI work in television and media commonly ask, in 2026, that every AI tool be properly licensed, that the license indemnify claims arising from the tool and its output, and that it be an enterprise license or equivalent so nothing put into the tool is used to train it or for anyone but the buyer.
The indemnity half is often negotiable. Buyers have accepted that AI tools do not indemnify claims about output made from material the buyer supplied, and have agreed that the vendor’s own AI media liability insurance, together with the vendor’s own contractual indemnity, satisfies the tool-indemnity requirement. That substitution should be written into the signed contract, not agreed on a call.
The rest of the package usually looks like this. You indemnify the buyer for your performance and for any breach or alleged breach of your warranties of originality, non-infringement, privacy and no defamation, softened to the best of your knowledge. You carve out footage and material the buyer supplied, edits the buyer made to the deliverables, and work done on the buyer’s instructions. Your total liability, including the indemnity, is capped, and the natural anchor for the cap is your E&O limit. Some buyers will also release the vendor from name, image, likeness and voice claims connected with the work. Many will not, but it costs nothing to ask.
Is the no-training requirement the same thing as the indemnity?
No, and it is the part insurance cannot replace. A clause that says the tool’s owner must not train on the buyer’s material, or use it for anyone else, is about data, confidentiality and the tool’s terms. Enterprise and business plans commonly promise that; consumer plans often do not. So a contract can still require an enterprise or equivalent plan for data reasons even after the buyer has accepted your insurance in place of the tool’s indemnity.
Treat them as two questions. Your media liability policy and capped indemnity can answer the indemnity question. The data question is answered by the plan you pay for, and by keeping the buyer’s unreleased material out of any tool whose terms let the vendor train on it.
How do you propose your insurance in place of the tool’s indemnity?
In the redline, in plain words. State that the AI tools used do not indemnify claims arising from output made from material the buyer supplies, and that the buyer agrees your media liability insurance covering AI-related claims, together with your indemnity under the agreement, satisfies the tool-indemnity requirement. Attach the certificate and offer the AI wording by form number.
Then line the numbers up. The cap on your liability should include the indemnity and should not exceed what your policy can pay. If the buyer wants a higher cap, the answer is a higher limit, priced before you agree to it.
Which AI platforms indemnify their output, and on which plans
Read from each vendor’s own terms or help pages on October 15, 2026. Terms change, enterprise contracts can differ from the published ones, and none of this is legal advice. "Could not confirm" means exactly that.
OpenAI (ChatGPT, API)
- Plan that carries an indemnity
- API, ChatGPT Enterprise, Edu and Business, under Copyright Shield. Not free or Plus.
- Main conditions and limits
- Excludes modified or combined output, known infringement, disabled safety or citation features, inputs you had no rights to, and trademark claims. Cap could not be confirmed.
OpenAI Sora
- Plan that carries an indemnity
- Could not confirm whether Copyright Shield reaches Sora output.
- Main conditions and limits
- Ask OpenAI in writing before relying on it.
Microsoft (Copilot, Azure OpenAI)
- Plan that carries an indemnity
- Paid commercial Copilot services and Azure OpenAI, under the Customer Copyright Commitment. Not free or consumer products.
- Main conditions and limits
- Safety systems and content filters must stay on. Azure OpenAI adds required mitigations and a testing report. Excludes known infringement, inputs you had no rights to, and trademark claims.
Google Cloud (Vertex AI, Gemini, Imagen, Veo)
- Plan that carries an indemnity
- Paid use of the listed generally available models, with two prongs, training data and generated output.
- Main conditions and limits
- Output must be unmodified. Excludes circumvented filters, use after a rights holder’s notice, and trademark claims. Preview models may not be listed.
Google Workspace (Gemini)
- Plan that carries an indemnity
- Paid Workspace plans, not credits or free tiers.
- Main conditions and limits
- The same two prongs and exclusions as Google Cloud.
Anthropic (Claude)
- Plan that carries an indemnity
- Paid commercial use, including the API and business plans. Consumer plans run on different terms.
- Main conditions and limits
- Excludes modified outputs, customer inputs, combinations and known infringement. Sits outside the liability cap.
Amazon (Bedrock, Nova, Titan, Amazon Q)
- Plan that carries an indemnity
- Amazon’s own listed models and services. Not other providers’ models on Bedrock, and not the Q Developer Free Tier.
- Main conditions and limits
- Excludes infringing inputs, fine-tuned or modified models, disabled filters and trademark claims. Stated not to be subject to a damages cap.
Adobe Firefly
- Plan that carries an indemnity
- Customers who buy the enterprise IP indemnity entitlement.
- Main conditions and limits
- Generally available image features. Excludes your edits, including in Creative Cloud. Capped by your Adobe contract. Firefly Video and partner models could not be confirmed.
IBM watsonx (Granite)
- Plan that carries an indemnity
- IBM-developed models, under IBM’s standard IP indemnity.
- Main conditions and limits
- IBM says uncapped for its own models. Third-party models may be capped.
Getty Images
- Plan that carries an indemnity
- Generative AI by Getty Images, with its license.
- Main conditions and limits
- Prompts naming real people, trademarks or protected works you lack rights to are excluded. Getty describes it as uncapped; check the contract you sign.
Shutterstock
- Plan that carries an indemnity
- Eligible enterprise plans.
- Main conditions and limits
- Follows the license limits. Trademarks, copyrighted works and public figures need Shutterstock’s human review first.
Canva (Magic Studio)
- Plan that carries an indemnity
- Eligible enterprise customers, under Canva Shield.
- Main conditions and limits
- Full terms and cap could not be confirmed.
Synthesia
- Plan that carries an indemnity
- Synthesia says it offers an IP indemnity for the content and technology it provides.
- Main conditions and limits
- Scope, plans and cap could not be confirmed.
Meta (Llama)
- Plan that carries an indemnity
- None. The license provides the models as is.
- Main conditions and limits
- The licensee indemnifies Meta.
Midjourney, Stability AI
- Plan that carries an indemnity
- None in the standard terms.
- Main conditions and limits
- The user indemnifies the vendor. A separate enterprise agreement could differ; not confirmed.
Runway, Luma Dream Machine, Pika, Kling
- Plan that carries an indemnity
- None in the standard terms.
- Main conditions and limits
- The user indemnifies the vendor. Enterprise terms, where they exist, could not be confirmed.
ElevenLabs, HeyGen
- Plan that carries an indemnity
- None in the standard terms.
- Main conditions and limits
- The user indemnifies the vendor. Enterprise agreements could differ; not confirmed.
Moonvalley (Marey)
- Plan that carries an indemnity
- None in its terms of service.
- Main conditions and limits
- Moonvalley describes Marey as trained on licensed data. Its terms still make the user responsible for outputs.
An indemnity protects the vendor’s own customer, the account holder. It does not reach your client unless you pass protection on through your own contract.
The detail
Microsoft’s Customer Copyright Commitment covers paid commercial Copilot services and Azure OpenAI, not free or consumer products, and only where the customer has not disabled or evaded the safety systems and content filters; Azure OpenAI customers must also keep the required mitigations and a testing and evaluation report.
Microsoft Learn, "Customer Copyright Commitment required mitigations" (updated July 13, 2026)
OpenAI’s Service Terms extend its copyright indemnity, which OpenAI calls Copyright Shield, to API customers and to ChatGPT Enterprise, Edu and Business, and exclude output the customer modified or combined with other products, output it knew or should have known was infringing, use with safety or citation features disabled, inputs it had no rights to, and trademark claims.
Google Cloud indemnifies paid use of its listed generative AI services twice, once for its use of training data and once for unmodified generated output, and the output promise does not apply where the customer disregarded or circumvented filters or citations, kept using output after a rights holder’s notice, or faces a trademark claim.
Google Cloud, Service Specific Terms, section 20(i) "Generative AI Indemnified Services"
Google Workspace gives the same two-part generative AI indemnity for Gemini in Workspace, but only where the customer pays for the service and the use is not under credits or a free tier.
Google Workspace, Service Specific Terms, section 12.10 (modified October 12, 2026)
Anthropic’s Commercial Terms defend paying customers against claims that their authorized use of Claude or its outputs infringes intellectual property, exclude modified outputs, customer inputs, combinations with other technology and known infringement, and place the indemnity outside the liability cap.
Anthropic, Commercial Terms of Service, section K (effective June 17, 2025)
Amazon’s generative AI indemnity covers its own listed models and services, including Amazon Nova, Titan and Amazon Q outside the Q Developer Free Tier, is stated not to be subject to any damages cap, does not extend to other providers’ models on Amazon Bedrock, and excludes infringing inputs, fine-tuned or modified models, disabled filters and trademark claims.
Amazon Web Services, Service Terms, section 50.10 (updated October 13, 2026)
Adobe’s Firefly IP indemnity is available to customers who buy the enterprise entitlement, covers generally available Firefly features that generate imagery, excludes the customer’s own modifications including edits made in Creative Cloud, and is limited by the cap in the customer’s Adobe contract.
Adobe, "Firefly legal FAQs for enterprise customers" (June 11, 2024)
IBM says it extends its standard intellectual property indemnity to output of IBM-developed watsonx models without a cap and does not ask customers to indemnify IBM for using them.
Shutterstock’s generative AI indemnity is limited to eligible enterprise plans, follows the limits of the license purchased, and covers output depicting a trademark, a copyrighted work or a public figure only after Shutterstock’s own human review approves it.
Meta’s Llama 4 Community License provides the models and their output as is, with no indemnity from Meta, and requires the licensee to indemnify Meta against third-party claims arising from its use or distribution.
Runway’s Terms of Use say use of outputs is at the user’s sole risk, give no indemnity for output, and require the user to indemnify Runway for the user’s content, which includes generated content; enterprise customers sign separate terms.
The standard terms of Midjourney, Stability AI, Luma Dream Machine, Pika, ElevenLabs, HeyGen, Moonvalley and Kling each disclaim that output is non-infringing, offer the user no indemnity, and ask the user to indemnify the vendor.
What this means for you
- Check your plan before you promise anything. If a contract asks for a tool indemnity, read your tool’s current terms for your plan. A free, consumer or creator plan very likely has none, and even an enterprise plan may not cover edited work or the client’s own footage.
- Buy the protection the tool will not give you. A media liability policy with affirmative AI wording covers your content whichever tool made it. It is the asset that makes your own indemnity worth something to a buyer.
- Put the substitution and the cap in the contract. Your insurance plus your capped indemnity in place of the tool’s, with the cap matched to your limit and the buyer’s own material carved out.
- Keep the data clause separate. No-training and confidentiality promises are about the plan you pay for. Insurance does not answer them.
Related questions
- Does ChatGPT Enterprise indemnify me for copyright claims?
OpenAI’s Service Terms extend Copyright Shield to API customers and to ChatGPT Enterprise, Edu and Business, not to free or Plus accounts. It excludes output you modified or combined with other products, output you knew was infringing, use with safety features off, and trademark claims. Whether it reaches Sora we could not confirm.
- Does Microsoft Copilot’s copyright commitment cover me?
If you use a paid commercial Copilot service or Azure OpenAI and leave the safety systems and content filters on, Microsoft’s Customer Copyright Commitment can defend you against copyright claims about the output. Free and consumer Copilot are not covered, and Azure OpenAI customers must also keep Microsoft’s required mitigations in place.
- Do Google Gemini, Vertex AI or Veo come with an indemnity?
Paid use of the generative AI services Google lists is indemnified twice, for Google’s training data and for unmodified generated output, on Google Cloud and in paid Google Workspace. Free tiers and credits are excluded, and preview models may not be on the list. Check that the model you use is listed before you rely on it.
- Does Adobe Firefly indemnify AI images and video?
Adobe sells an IP indemnity to enterprise customers for generally available Firefly features that generate imagery, capped by the customer’s Adobe contract. Edits you make afterward, including in Creative Cloud, fall outside it. We could not confirm from Adobe whether it reaches Firefly Video or the partner models inside Firefly.
- Do Runway, Midjourney or ElevenLabs indemnify their output?
Not in their standard terms. Runway, Midjourney, Luma Dream Machine, Pika, Kling, ElevenLabs, HeyGen, Stability AI and Moonvalley each disclaim that output is non-infringing and ask the user to indemnify them. An enterprise agreement may differ, so read your own contract if you have one.
- Do I need an enterprise AI plan to meet a client’s indemnity clause?
Not necessarily. Buyers commonly accept your own media liability policy that covers AI content, plus your own capped indemnity, in place of the tool’s indemnity. A clause about not training on the client’s material is different, and may still require an enterprise or equivalent plan.
- The AI clauses in a client contract, one by one
- Will my insurance pay what the AI vendor would have paid?
It pays covered claims against you, subject to the policy wording, limit and deductible. It does not stand in the vendor’s shoes or make the vendor pay anyone. Check that the policy is affirmative or silent on AI rather than excluding it, and whether it covers liability you take on in an indemnity.
- What to check on an AI media liability policy
- Does an enterprise indemnity protect my client?
No. A tool indemnity protects the vendor’s customer, the account holder. If you hold the account, your client is protected only by what you promise in your own contract, which is why your insurance and your indemnity matter even when your tools do indemnify you.
Need this handled?
Reading about it and having it handled are different jobs, and the second one is ours. We place the accounts other brokers decline — claims history, a cancellation, an unusual operation, a business nobody wants to underwrite twice — and we will tell you when the policy you already have is the right one. Tell us the situation and a licensed human replies the same business day.
Written by Kevin Kelsey, founder and principal agent, CA License #6011822, at Boring Insurance. Last updated 2026-10-15. See all guides.