AI consulting contract: the clauses to check before you sign
This is a buyer's checklist, not legal advice, and your own lawyer should read the final text. It covers ten clauses worth checking and credits each legal point to the page it comes from.
AI consulting contract
- Clauses covered
- Ten, in a table
- GDPR anchor
- Article 28, processor contracts
- AI Act anchor
- Provider and deployer duties
- Checked
- 12 November 2026
- Status
- Buyer checklist, not legal advice
The short answer
An AI consulting contract should state the scope and how you accept it, who owns the code and evaluations, how personal data and subprocessors are handled under GDPR Article 28, which EU AI Act role each party holds, liability, termination and handover. Ask for each in writing before work starts, and have a lawyer read the result.
What should an AI consulting contract cover?
The same things any software contract covers, plus four that AI work sharpens. Model providers sit behind the consultant as subprocessors. Output is probabilistic, so acceptance needs a test, not a demo. Evaluation sets and prompts are valuable and easy to forget in an ownership clause. And the EU AI Act gives different duties to different roles.
A contract is also the cheapest place to find out whether a firm is serious. If a consultant resists writing down who owns the code or which subprocessors touch your data, you have learned something before spending anything. Pair this page with the AI vendor due diligence checklist and the firm versus freelancer comparison, since freelancers need the same clauses.
None of this is legal advice. Treat it as the list of things to raise with your lawyer.
Which ten clauses matter most?
| Clause | What to ask for | Red flag |
|---|---|---|
| Scope and deliverables | A written list of deliverables per milestone, and what is out of scope | Scope described as an outcome with no list |
| Acceptance | Acceptance tests you can run, with pass criteria agreed in advance | Acceptance by demo or by silence |
| Code and IP ownership | Code, prompts, evaluation sets and configuration are yours from the first commit | A licence back to you for your own system |
| Personal data and processor terms | A data processing contract that follows GDPR Article 28 | No mention of data protection at all |
| Subprocessors and model providers | A named list and prior written authorisation for changes | Providers added without notice |
| Security | Work in your accounts with least-privilege access and an access removal date | Production data copied to personal machines |
| EU AI Act roles | A statement of who is provider and who is deployer, and who handles incidents | Silence on roles |
| Liability | A cap you have discussed and carve-outs for data breach and IP | A cap equal to one month of fees, unexplained |
| Termination | Exit at a cycle boundary with all work delivered up to that date | Long lock-in with no exit for cause |
| Handover | Runbook, walkthrough and one change shipped by your own engineer | Handover as a slide deck |
What does GDPR Article 28 require of a consultant who touches personal data?
Sufficient guarantees. Article 28(1) says the controller uses only processors providing sufficient guarantees to implement appropriate technical and organisational measures. In practice you should be able to ask a consultant what those measures are and get a written answer.
Subprocessors. Article 28(2) says the processor shall not engage another processor without prior specific or general written authorisation of the controller. Model providers and hosting vendors count, so the contract should name them and require notice before a change.
A binding contract. Article 28(3) says processing by a processor is governed by a contract or other legal act binding the processor, setting out the subject matter and duration. A statement of work with no data clause does not meet that description on its face.
Your role. Whether the consultant is a processor depends on the facts. If it only touches data on your instructions, the Article 28 terms apply. Ask your lawyer to confirm the roles in the specific project.
Which EU AI Act role is the consultant in, and which are you in?
The European Commission's AI Act page sets out the duties by role. We state only what it says.
- It says providers must keep a post-market monitoring system in place, while deployers handle human oversight and monitoring.
- It says both providers and deployers must report serious incidents and malfunctions.
- It says the AI Act defines four levels of risk, from unacceptable to minimal, and that it became applicable on 2 August 2026, with exceptions.
- It says providers of generative AI have to ensure that AI-generated content is identifiable.
For a custom build, the practical questions are who builds the system, who puts it into use, and who reports an incident. Our provider or deployer guide walks through the roles. Write the answer into the contract so nobody discovers it after an incident.
Who owns the code, prompts and evaluations?
You should, from the first commit, and the contract should say so in words. The G-Cloud 15 listing for Elsewhen's forward deployed AI engineering service lists freedom from vendor lock-in through full ownership of IP as a benefit, so a buyer can reasonably ask for it. Winder.AI's buyer guide asks who owns the code and pipelines at the end of the engagement.
Be specific. Name the repository, the infrastructure, the prompts, the evaluation datasets, the fine-tuned weights if any, and the documentation. State that each lives in accounts you control, and that the consultant keeps no copy after handover beyond what the contract allows. Ask separately about tools the consultant brings, such as reusable libraries, and get a licence to keep using them.
Third-party models and open-source components carry their own licences. List them in an appendix so that your lawyer can check them once.
What should the liability, security and termination clauses say?
These three are where negotiation time is well spent, and none of them needs legal jargon to raise.
- Liability. Ask what the cap is, what it is based on, and which events sit outside it, such as a data breach or an IP claim. A cap that only covers fees paid is a decision, so make it consciously.
- Security. Require work in your accounts, least-privilege access, secrets in your own manager, a list of who can reach production data, and a date on which access is removed.
- Incidents. Agree who tells whom, and within how long, if a model leaks data or an agent takes a wrong action. Link this to the incident duties in the AI Act section above, and test it once with a tabletop exercise before you go live.
- Termination. Prefer exit at a cycle boundary, with everything built so far delivered to your repository. Add exit for cause if the firm misses agreed milestones twice.
- Change control. Agree how a new requirement is estimated and approved, so that a request in week four does not become an argument.
How do you review the contract in four steps?
Do it in this order and the review takes days, not weeks.
- Mark the ten clauses in the table above. Highlight any that are missing, not just any you dislike.
- Send the list of gaps to the consultant and ask for a revised draft, not a call.
- Hand the revised draft to a lawyer with three questions: data protection roles, liability cap, and IP. See the scope of work checklist for the scope side.
- Agree a handover test in advance. For example, one of your engineers ships a small change using only the runbook.
Where Digiton fits, and what to check on any firm
Digiton Dynamics is an AI consulting firm that builds AI agents and custom AI software, and takes systems into production with production engineering. Its production review gives you a scorecard across ten layers with evidence for each. Digiton Dynamics OÜ is registered in Tallinn, Estonia, so the contracting entity is Estonian. Ask your lawyer what that means for governing law. Ask any firm, Digiton included, which entity signs and where it is registered.
Digiton is not a law firm and this page is not legal advice. Read the contract, ask for each clause in the table, and ask the same of every firm. Cost sits in a separate guide, AI consulting cost.
Frequently asked questions
What should be in an AI consulting contract?
Scope and acceptance tests, ownership of code, prompts and evaluations, a GDPR Article 28 data processing contract, a named subprocessor list, AI Act roles, a liability cap, termination terms and a handover.
Does a consultant need a data processing agreement?
If the consultant processes personal data for you, Article 28(3) says processing is governed by a contract binding the processor. Ask your lawyer to confirm the roles for your project.
Who should own the code an AI consultant writes?
You, from the first commit, with the contract naming the repository, prompts and evaluation sets. Some listings, such as Elsewhen's on G-Cloud 15, offer full ownership of IP as a benefit.
Is this legal advice?
No. It is a buyer checklist. Have a lawyer read the final contract.
Sources, checked 2026-11-12
6 sources
- GDPR Article 28(1) says the controller uses only processors providing sufficient guarantees to implement appropriate technical and organisational measures.Source: eur-lex.europa.eu, checked 2026-11-12.
- GDPR Article 28(2) says the processor shall not engage another processor without prior specific or general written authorisation of the controller.Source: eur-lex.europa.eu, checked 2026-11-12.
- GDPR Article 28(3) says processing by a processor is governed by a contract or other legal act binding the processor, setting out the subject matter and duration.Source: eur-lex.europa.eu, checked 2026-11-12.
- The European Commission's AI Act page says providers must keep a post-market monitoring system while deployers handle human oversight and monitoring, that both must report serious incidents, that the Act defines four levels of risk, and that it became applicable on 2 August 2026 with exceptions.Source: digital-strategy.ec.europa.eu, checked 2026-11-12.
- The Elsewhen Ltd Forward Deployed AI Engineering Service is listed on G-Cloud 15 and lists freedom from vendor lock-in through full ownership of IP as a benefit.Source: applytosupply.digitalmarketplace.service.gov.uk, checked 2026-11-12.
- Winder.AI's 12 August 2026 buyer guide suggests asking who owns the code and pipelines at the end of an engagement.Source: winder.ai, checked 2026-11-12.
Digiton Dynamics OÜ is registered in Tallinn, Estonia, and delivers from Lisbon.
Update log
- Published. Read the GDPR text on EUR-Lex (Article 28), the European Commission AI Act page, the Elsewhen G-Cloud 15 listing and Winder.AI's buyer guide on 12 November 2026.
Each source above was opened on the date beside it. How facts get chosen and checked is written down in the editorial policy.
Talk to an engineer about your system
Book a free 25-minute scoping call. We tell you what we would build, what it takes to keep it running, and whether we are the right fit.