Three questions decide most of an enterprise AI contract: how long the vendor keeps your data, whether it may train on it, and who owns the outputs and any model tuned on your material. Settle each in writing before you sign, because a sales call and a contract can say different things.

1. Retention: what is kept, where and for how long

  • Prompts and files. Ask whether inputs are stored at all, for how long, and whether the period differs for logs, backups and abuse monitoring.
  • Deletion. Ask how you delete data, how fast deletion reaches backups, and whether you get written confirmation.
  • Location and access. Ask where data is held and which of the vendor’s staff or sub-processors can see it.

2. Training and fine-tuning: whose data teaches the model

  • Training on your data. Ask for a clear yes or no on whether your inputs and outputs are used to improve the vendor’s general models. Look for the answer in the contract, not only in a policy page that can change.
  • Fine-tuning. If the vendor tunes a model for you, ask where the tuned weights live, who can use them, and what happens to them when the contract ends.
  • Provenance. Ask how the vendor shows that its own training data was licensed. We wrote about this in enterprise AI training data and model licence provenance.

3. Ownership: outputs, tuned models and your records

  • Outputs. Ask who owns what the system produces, and whether anything limits how you use it.
  • Tuned models. Ask whether a model tuned on your data is yours, licensed to you, or the vendor’s.
  • Exit. Ask whether you can export your records, prompts, evaluation results and configuration when you leave, and in what format.

What to put in writing

  1. A retention period for each data type, and a deletion deadline after termination.
  2. A statement that your data is not used to train general models, or the exact exceptions.
  3. Who owns outputs and tuned models, and what happens to both on exit.
  4. Notice before any change to these terms, and a right to leave if you do not accept it.

This is a buyer’s checklist, not legal advice. Terms differ by vendor and by contract, so have counsel read your own. For what happens after signing, see how model version changes can drift from what you validated.

Where this meets capital

Proof of Talk is an invitation-only executive summit on capital, digital assets and AI. Abu Dhabi, on 3–4 December 2026 at the Louvre Abu Dhabi, is capped at 2,000 people; Paris, on 2–3 June 2027 at the Musée des Arts Décoratifs, Louvre Palace, is capped at 2,500. Admission is by application and review. The summit’s subject is the pairing of capital and AI; our guide to the capital and AI pairing explains why, and the Abu Dhabi edition page has the programme.

The FAQ covers admission. To apply, request access to Abu Dhabi. A person reviews each request and it is not a guarantee of a place.

How this guide was made

Proof of Talk’s editorial team wrote it on 19 November 2026 from general knowledge of how enterprise software contracts are structured and from our own earlier guides. It is a checklist of questions, not a reading of any vendor’s terms. We organise the events it mentions, so it is not a neutral comparison. It quotes no figure we cannot source and makes no claim about any other event.