Eliga · Article header hero (EH10 · responsive)
AI Risk & Commercial Contracts Insights 3 min read

Three AI Clauses Hiding in Your 2026 Vendor Contracts

The fastest-moving companies are signing them without noticing. Here is what to look for before you do.

Book a contract review Read the article
Eliga · Three AI Clauses Hiding in Your 2026 Vendor Contracts

Three AI Clauses Hiding in Your 2026 Vendor Contracts

The fastest-moving companies are signing them without noticing. Here is what to look for before you do.

Eliga Consultancy
Insights · 3 min read

Most founders skim a vendor contract for price, term, and termination. In 2026, that is no longer enough. As suppliers bake AI into their products, they are quietly rewriting the risk terms, and the default position increasingly favours them, not you. Three clauses in particular are worth stopping on.

The “AI output” disclaimer

Many vendors now include wording that disclaims all responsibility for anything their AI feature produces. On paper it sounds reasonable. In practice, if that output feeds into your product or your decisions and it is wrong, the liability has been routed to you. The commercial question is simple: if their tool causes the harm, who explains it to your customer? Right now, the contract often says you do.

The data and training grant

Watch for clauses giving the vendor rights to use “your data” to “improve their services.” That phrasing can quietly permit them to train their models on your confidential information and your customers’ data. Beyond the obvious privacy exposure, you may be handing a supplier the right to learn from the very data that makes your business defensible.

The shifting indemnity

Traditionally the vendor indemnifies you if their product infringes someone’s rights. New AI clauses increasingly carve out AI-generated output from that protection, or flip it, so that you indemnify them against claims arising from how you used the AI. The protection you assume you have may have been removed in a single sub-clause.

You may be handing a supplier the right to learn from the very data that makes your business defensible.

None of this means avoiding AI vendors. It means reading the risk terms with the same care you give the price. The companies getting caught are not careless. They are simply moving fast and trusting that a standard contract still means what it used to.

Common questions

AI clauses, answered

What AI clauses should I watch for in vendor contracts?
Three in particular: an AI output disclaimer that routes liability for the AI’s output onto you; a data and training grant that lets the vendor train its models on your data; and a shifting indemnity that carves AI-generated output out of the protection you would normally expect, or reverses it so you indemnify the vendor.
Who is liable when a vendor’s AI feature gets it wrong?
It depends on the contract, but AI output disclaimers increasingly push that liability onto the customer. If the AI output feeds your product or decisions and is wrong, you may be the one who has to answer to your own customer, even though the vendor supplied the tool.
Can a vendor train AI on my data?
If the contract grants rights to use your data to improve their services, often yes. That wording can permit training on your confidential information and your customers’ data, so it should be reviewed before signing.

This article is general information on commercial and AI contract risk, and is not legal advice.