Six Clauses to Negotiate in Any AI Contract

In an AI contract, price is usually the least negotiable term and among the least important. The six worth spending your leverage on are data rights, exit, price protection, service levels, model-change notice and indemnity — and all six are far cheaper to secure before signature than to discover at renewal.

The six

1. Data rights. Does your data train their models, and can you turn that off? Get the answer in the contract rather than from a salesperson, and check the default rather than the option. The UK ICO's guidance on AI and data protection is a useful reference if you handle personal data.

2. Exit. What you get on the way out, in what format, and how long they keep it afterwards. Also what happens to anything derived from your data. A vendor without a clear answer has built lock-in whether or not they meant to.

3. Price protection. A cap on renewal increases. AI pricing has moved sharply in both directions, and an uncapped renewal on a system you have integrated deeply is an unpriced risk.

4. Service levels that mean something. Uptime is easy to promise. The harder and more useful commitment is latency and, where relevant, an accuracy floor on your data — with a remedy attached, or it is decoration.

5. Model-change notice. The vendor will swap the underlying model. When they do, your evaluation results are void. Notice plus a window to re-test is reasonable to ask for and rarely offered unprompted.

6. Indemnity. Who is liable if the output infringes something or causes loss. Read what is actually covered rather than the headline that indemnity exists.

Where your leverage actually is

Before signature, and nowhere else. Once the tool is integrated and your team depends on it, every one of these becomes a request rather than a negotiation.

This is why the evaluation and the contract should run together rather than sequentially — the point of the vendor evaluation framework, and why vendor red flags include a refusal to discuss exit terms.

The asymmetry worth remembering

The vendor has negotiated hundreds of these; you have negotiated a few. They know which clauses they concede routinely, and you do not. That asymmetry is closed by asking directly which of these six they have granted before — a question that is difficult to dodge and surprisingly effective.

None of this is legal advice, and anything material should go past counsel. What a Chief AI Officer owns is knowing which terms to send them — the accountability described in what does a Chief AI Officer do.

Frequently asked questions

What should I negotiate in an AI contract? Data rights, exit terms, renewal price caps, meaningful service levels, notice of model changes, and indemnity. Price itself is usually the least flexible and least consequential term.

Why do model-change terms matter? Because when a vendor swaps the underlying model, every evaluation you ran is void. Notice and a window to re-test protects you from silently degraded performance.

When do we have leverage? Only before signature. After integration, each of these becomes a request rather than a negotiation, which is why evaluation and contracting should run in parallel.

Do we need a lawyer? For anything material, yes. The role of an AI leader is knowing which terms are worth counsel's time, not replacing counsel.

Further reading

• The AI vendor evaluation framework
• Nine AI vendor red flags
• Build vs buy AI
• The Fractional Chief AI Officer retainer