Industry Insights
Words of wisdom from our business insurance experts.
Who's Liable When Your AI Product Gets It Wrong?

Companies that sell or deploy AI products can face claims even when the underlying error originates in a third-party model. Who is legally responsible depends on the alleged harm, each party's role, applicable law and the relevant contracts. Technology errors and omissions (Tech E&O), cyber, media liability and directors and officers (D&O) insurance may each respond to different AI claims, but coverage depends on policy wording, and some insurers are now adding AI-specific exclusions.
This guide is for US startups building AI into their products or using it to serve customers. It covers who can be held responsible, which policies may respond, where exclusions and coverage gaps may arise, and how contracts and good practice can reduce your exposure.
Who can be held responsible
AI rarely creates a new kind of liability. Instead, familiar claims, such as negligence, misrepresentation, breach of contract, IP infringement, privacy violations and discrimination, arise in new ways. Three parties can be pulled in:
- The company that sells or deploys the AI product. This is usually you. Customers and the public deal with your product, your brand and your contract, so claims can start there. Being named in a claim doesn't mean you're liable, but you'll still need to respond to it.
- Your customer, if it uses your AI to serve its own users. A customer facing a claim may then turn to you under the indemnity in your contract.
- The model or platform provider. If the error came from a third-party foundation model, you may want to pass the loss up the chain. In practice, provider terms often limit their liability, so recovery can be hard.
A US case shows how vendors can be drawn in. In Mobley v. Workday, pending in federal court in California, job applicants allege that the AI screening tools in Workday's hiring platform discriminated against them. Workday argued it wasn't the employer. The court allowed certain claims to proceed on the theory that Workday could be treated as an agent of the employers using its software, and in May 2025 preliminarily certified a collective action on the age discrimination claim. These procedural rulings do not establish that Workday discriminated. They illustrate why an AI vendor cannot assume that its customer's role in the final employment decision eliminates the vendor's exposure.
Common AI claim scenarios and which policy may respond
Which policy responds depends on what's alleged, not on the fact that AI was involved. These are typical scenarios and the coverage most likely to be relevant:
For software companies selling AI products, financial-loss and privacy claims are often the most relevant exposures, which is why Tech E&O and cyber are usually the core coverages to review. For more on these lines, see Tech E&O Insurance: Why Software and IT Companies Can't Afford to Skip It.
Where coverage stops: AI exclusions and affirmative AI coverage
Some policies do not expressly address AI. This is sometimes described as "silent AI," but silence does not establish coverage: existing insuring agreements, definitions, conditions and exclusions still determine whether a claim is covered. Some insurers are now addressing AI directly:
- Standard general liability exclusions now exist. Verisk's ISO has released generative AI endorsements that insurers can add to general liability policies, including CG 40 47, CG 40 48 and CG 35 08, which remove coverage for certain injury, damage or personal and advertising injury tied to generative AI, according to Insurance Journal. AI exclusion endorsements are available and are appearing on some policies. Their effect depends on the wording and whether they are actually part of your policy.
- Some insurers have broader exclusions. The same report describes an absolute AI exclusion introduced by one insurer for D&O, E&O and fiduciary liability products, reaching liability tied to the use or development of AI and AI-generated content.
At the same time, some insurers offer coverage written specifically for AI risks, either as endorsements to Tech E&O and cyber policies or as standalone products. Terms vary widely. Ask whether the proposed coverage addresses the specific AI activities you perform, the losses you could face and the contractual obligations you accept, not merely whether the product is marketed as "AI insurance."
What to do at renewal: ask your broker to list every AI-related exclusion, limitation or definition across your policies, and to compare how different insurers' wording treats the AI activities you actually perform. An exclusion added at renewal can remove the coverage you most need.
How contracts shift AI risk
Your contracts often decide who pays for an AI error before any insurer is involved. Four places to look:
- Warranties and performance promises. Promising that your AI will be accurate, unbiased or fit for a specific purpose can create liability beyond ordinary professional care, which insurers may treat as contractual liability. Describe what the product does carefully and avoid guarantees you can't control.
- Customer indemnities. Enterprise customers increasingly ask for indemnities covering AI outputs, IP infringement from generated content, or misuse of their data in training. Check whether your insurance would respond to what you're agreeing to.
- Limitation of liability. A cap tied to fees can protect you, but customers may push to carve AI-related claims out of the cap. Negotiate defined caps or super-caps rather than unlimited exposure.
- Model and platform provider terms. Read what your AI providers promise you. Many limit their liability and offer narrow indemnities, if any. Any gap between what you promise customers and what providers promise you is your exposure.
Regulation to watch
AI rules are still taking shape, and they can create new duties, and new grounds for claims, for companies that build or use AI. This is a brief overview, not a compliance guide; check the official text of any law that may apply to you.
- Federal. There's no single federal AI law, but the FTC has continued to act against companies that make false or unsupported claims about what their AI can do, according to Holland & Knight. A December 2025 executive order directs challenges to state AI laws, but an executive order doesn't by itself repeal them.
- Selected state and local requirements. A growing number of state and local rules address AI, often in hiring, consumer disclosures and automated decisions. Examples include New York City's bias audit rules for automated hiring tools, and California's CCPA regulations, which took effect January 1, 2026, with compliance for covered automated decision-making technology requirements beginning January 1, 2027. Applicability depends on whether the business and its use of the technology fall within the rules. Colorado enacted SB26-189, replacing its earlier AI framework with requirements addressing automated decision-making in consequential decisions, according to Eckert Seamans. The statute specifies January 1, 2027, but litigation has affected enforcement: Kelley Drye describes a federal court's April 2026 order staying enforcement in xAI v. Weiser. Confirm the current court orders and implementing rules before relying on that date as an enforcement deadline. For a broader state-by-state summary, see Security Boulevard.
Regulatory investigations and fines are treated differently from lawsuits in many policies, and fines may be uninsurable in some places. If you sell into regulated industries, or your AI is used in hiring, lending, housing or health care, ask your broker how your D&O, cyber and Tech E&O policies treat regulatory matters involving AI.
Practical steps to reduce AI liability exposure
1. Map where AI touches customers. List the features, decisions and outputs that rely on AI, and which ones could cause financial loss, privacy issues or unfair outcomes.
2. Keep a human in the loop where stakes are high. For decisions about money, health, employment or legal rights, add review steps or clear limits on what the AI can do on its own.
3. Set expectations in writing. Describe what your AI does and doesn't do in your terms, documentation and product. Avoid marketing claims you can't support.
4. Keep proportionate records. Maintain records of model versions, testing, material outputs, human review and incidents. Set retention periods and access controls, and avoid retaining personal, confidential or customer data unnecessarily.
5. Review your provider contracts. Know what your model and platform providers promise, and where their responsibility ends.
6. Align contracts and insurance. Have your broker compare the AI-related promises in your customer contracts with what your Tech E&O, cyber and other policies actually cover.
7. Review coverage at every renewal. Watch for new AI exclusions and changed definitions, and ask about affirmative AI coverage.
Frequently asked questions
Who is liable when an AI product makes a mistake?
Companies that sell or deploy AI products can face claims even when the underlying error originates in a third-party model. Who is legally responsible depends on the alleged harm, each party's role, applicable law and the relevant contracts with customers and model providers. Being named in a claim isn't the same as being liable.
Does Tech E&O insurance cover AI errors?
It may, if the claim alleges your product or service failed and caused a financial loss, and the policy doesn't exclude AI-related claims. Check for AI exclusions, contractual liability limits and how the policy defines your services.
Can I blame the AI model provider?
Do not assume your model provider will reimburse the loss. Review the applicable liability caps, indemnities, exclusions and claim procedures with counsel, and assess the exposure that remains with your company.
What is an AI exclusion in an insurance policy?
A provision that removes coverage for some or all claims involving artificial intelligence. Standard generative AI exclusions are now available for general liability policies, and some insurers have added broader AI exclusions to management and professional liability policies.
Is there insurance specifically for AI risks?
Some insurers offer coverage written specifically for AI, either as endorsements to Tech E&O or cyber policies or as standalone products. Terms vary widely, so compare definitions, triggers and exclusions carefully.
Which US AI laws apply to startups?
There's no single federal AI law, but federal consumer protection rules still apply, and a growing number of states and cities have AI rules, including New York City and California. Colorado's replacement law specifies January 1, 2027, but litigation has affected enforcement, so confirm its current status. Which rules apply depends on where you operate, who your customers are and how your AI is used, so get legal advice.
Talk to a Fullsteam advisor
Building AI into your product? Talk to a Fullsteam advisor
This article is general information, not insurance, legal or financial advice. Coverage, exclusions and availability vary by insurer, policy wording and the facts of each claim, and AI law is changing quickly. Have your broker and counsel review your situation.
management specialist





