← All insights Contracting

5 AI-specific clauses every technology contract now needs

Most technology contracts in circulation today were drafted for a world of deterministic software — software that does the same thing every time you run it. AI does not behave that way. It is trained on data, it produces probabilistic outputs, it can change as the model is updated, and it raises questions about who owns what comes out the other end. Standard SaaS templates simply do not allocate those risks.

Whether you are buying an AI tool, embedding a third-party model, or selling an AI-enabled product, here are five clauses that should now be in the contract — and the questions each one needs to answer.

1 Data use and training rights

The most valuable — and most overlooked — question in any AI contract is: what can the provider do with your data? Specifically, can your inputs, prompts or documents be used to train or improve the provider's models? For many organisations the answer must be "no," particularly where the data is personal, confidential, or commercially sensitive.

If the contract is silent on training rights, assume the worst and negotiate for the position you actually want.

Address: permitted purposes for your data, an explicit prohibition (or carefully scoped permission) on training, data retention and deletion, and whether any data leaves the EEA.

2 IP and ownership of outputs

Who owns what the AI generates? The legal position on AI-generated output is unsettled in many jurisdictions, so the contract has to do the heavy lifting. You want clarity on ownership or a broad licence to use the outputs, plus warranties that using those outputs will not infringe third-party rights.

Address: assignment or licence of outputs to you, the scope of that licence (commercial use, modification, sublicensing), and an IP-infringement indemnity covering both the model and its outputs.

3 Model provenance, performance and change

AI providers update models — sometimes silently. A model that performed well at signing can behave differently after an update. Your contract should give you visibility and, where it matters, control.

Address: disclosure of the underlying model(s) and any sub-processors, notice of material model changes, performance/accuracy commitments where feasible, and the right to test or evaluate before changes take effect in production.

4 Liability, indemnities and allocation of AI risk

When an AI system produces a harmful, discriminatory or simply wrong output, who bears the consequences? Providers will push to cap or exclude liability; deployers need enough protection to cover regulatory exposure and downstream harm. This is where AI contracts are won or lost.

Address: indemnities for IP infringement and for breaches of data-protection law, carve-outs from liability caps for the risks you cannot absorb, and a clear allocation that reflects who actually controls the system.

5 Regulatory compliance and cooperation

The EU AI Act imposes obligations on both providers and deployers, and compliance often depends on information only the other party holds. The contract should require each side to provide what the other needs — technical documentation, logs, instructions for use — and to cooperate on regulatory requests.

Address: warranties of compliance with the EU AI Act and GDPR, an obligation to supply the documentation needed for your own compliance, cooperation on regulator or audit requests, and allocation of responsibility for conformity assessments where relevant.

The through-line

Each of these clauses is really about the same thing: putting the risk where it belongs. AI shifts risk in ways legacy templates never anticipated — into data, into outputs, into models that change underneath you. A contract that names those risks and allocates them deliberately is worth far more than one that hopes the old boilerplate still fits.

This article is provided for general information only. It is not legal advice and does not create a lawyer–client relationship. Contract terms must be tailored to the specific deal and governing law; obtain qualified advice for your circumstances.