Vendor Lock-In in AI Contracts: What to Check Before You Sign

You don't get locked into an AI vendor on day one. It happens gradually, over eighteen months of workflows built around their interface, staff trained on their quirks, and data accumulating in their format. By the time the contract comes up for renewal, walking away looks harder than it did when you signed.

I've reviewed enough AI vendor contracts to see the pattern. The demo focuses on capability. The sales conversation focuses on price. Nobody at the table asks what happens if you want to leave in two years. That question gets answered by the contract, whether you read it or not.

Every AI contract should say, in plain terms, what data you get back if you leave, in what format, and within what timeframe. Look for a specific number of days, not 'reasonable efforts'. Look for a named file format you can actually use elsewhere, not a promise of 'standard export'.

Government procurement guidance for AI now expects exactly this kind of clarity from suppliers, including proof that data can be extracted and reused outside the platform. If a supplier bidding for public sector work is expected to meet that bar, there's no reason your business should accept less.

If the contract is silent on data return, or vague about the format, treat that as your answer. A vendor confident in its product doesn't need to make leaving difficult.

AI vendors update the models running underneath their product. What you bought as a tool built on one model may be running on a different one within months, changing how it behaves, what it outputs, and how reliable it is for your use case. Most contracts don't mention this at all.

Ask for advance written notice before any material model change, and get 'material' defined against your use case, not the vendor's internal roadmap. Ask whether you get a testing window before a forced upgrade goes live. If the answer is no, you're accepting that your supplier can change the product you rely on without telling you.

A contract that protects you on exit has four things in writing: a data return format you can use, a fixed timeframe for that return, written confirmation of deletion once you've confirmed receipt, and a level of transition support that isn't left to goodwill.

None of this is exotic. It's the same due diligence you'd apply to any supplier holding critical business data. AI contracts get treated differently because the sales process moves fast and the technology feels new. The commercial risk isn't new. It's the same risk it's always been: what happens when this relationship ends.

Pull your current AI vendor contracts and check three things this week: whether the data export clause names a format and a deadline, whether there's a notice period for model changes, and whether exit support is written down or left to chance. If any of those three are missing, go back to the vendor and ask for them in writing before you renew. If you're about to sign a new contract, don't let the exit terms be the last thing you look at. Make them one of the first.