Clients have started writing AI use into freelance contracts, and most of them do it with one blanket line: no AI tools without written consent. Don't sign that blindly, and don't quietly ignore it either. Propose a short AI annex instead: which tools, which data may go into them, where they run, who owns the output and what you still review yourself. A clause you can follow protects you better than a ban you will break on the first day.
I see these clauses more often now in contracts and in the questionnaires that come with a request for proposal. They usually arrive late, in the last round of contract edits, written by someone in purchasing who copied them from another contract. That is the moment to answer them properly.
Why the blanket ban is a problem for both sides
A blanket ban sounds safe to the client. In practice it is either unenforceable or it puts the freelancer in breach from day one. Most IDEs now ship an assistant, the search engine summarises with a model, and the translation tool is a model too. Read strictly, "no AI tools" covers all of that.
The bigger risk is not the tool. It is the AI use that nobody writes down. When a contractor uses tools the client doesn't know about, client information can end up in places the client never approved. I covered the employee version of this in shadow AI and staff who hide their AI use. For a freelancer the same behaviour touches confidentiality, data protection and, at the end, the client's own legal position.
The fix is boring: write down what you actually do, and agree on it.
The five clause types I see
- The blanket ban or consent clause. "The contractor shall not use AI tools without prior written consent."
- Confidentiality. No client data, code or documents in external AI services.
- Data protection. Personal data only in tools covered by a processing agreement (Auftragsverarbeitungsvertrag, AVV).
- IP and output. Work results must be transferable to the client, and AI-generated parts must be disclosed.
- Competence. The contractor must know how to use the tools responsibly.
None of these is unreasonable on its own. Each one protects something specific. The trouble starts when a single line tries to cover all five.
The legal anchors behind them
I'm not a lawyer and this is not legal advice. But it helps to know which rules the client's legal team is thinking of, because that tells you what the clause is really for.
Trade secrets. Under the German Trade Secrets Act, information only counts as a trade secret if it is "Gegenstand von den Umständen nach angemessenen Geheimhaltungsmaßnahmen" by its owner (GeschGehG § 2 Nr. 1 b). A client who lets its information flow into tools nobody controls weakens its own position if it ever has to defend that secret.
Personal data. GDPR Article 28 requires a controller to "use only processors providing sufficient guarantees", and "Processing by a processor shall be governed by a contract or other legal act". That contract has to be in writing, which includes electronic form. If personal data goes into an AI service, someone has to have that contract with the provider.
AI literacy. The EU AI Act, Article 4, requires providers and deployers of AI systems to ensure, to their best extent, "a sufficient level of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf". It applies since 2 February 2025. A freelancer working for the client falls under "other persons" quite naturally, which is where the competence clause comes from.
Copyright. Under UrhG § 2 Abs. 2, "Werke im Sinne dieses Gesetzes sind nur persönliche geistige Schöpfungen." Output that a machine produced without a personal creative contribution may not be protected at all. A client paying for exclusive rights has a reason to ask what part of the deliverable is yours.
How I answer an AI clause
1. Ask what the clause is protecting
Before negotiating wording, ask the client which of the four risks they mean: secrets, personal data, IP or competence. The answer is often narrower than the clause. A bank that worries about customer data has no problem with a model that rewrites your documentation, as long as no customer data is in it.
2. Offer an AI annex instead of a yes or no
The annex is one page and replaces the blanket line. It lists:
- the tools, with the vendor and the data region;
- the data classes: public, internal, and never (personal data, credentials, customer data);
- where each tool runs: in the client's tenant, as a local model, or on the freelancer's own account;
- who owns the output, and that the freelancer reviews everything before delivery;
- disclosure of AI use on request;
- how a new tool gets added.
3. Run confidential work where the client can see it
For anything confidential, use the client's own AI environment or a model that runs locally. That takes the data-leakage question off the table. Many enterprise clients already have a licensed assistant in their tenant; ask for access before you reach for your own.
4. Keep a short record per deliverable
One line per deliverable is enough: which tool, for what, reviewed by you. It costs two minutes and answers the disclosure question before anyone asks it.
5. Price the ban if they insist on it
If a client bans every tool, the estimate changes. Say so openly in the offer. I made the related point about rates in AI discounts in time-and-material contracts: if the client wants the efficiency, they share it; if they don't want the tools, they pay for the hours.
An example annex clause
For illustration only, not legal advice:
The Contractor may use the AI tools listed in Annex [X] for the services
under this agreement. The Contractor shall not enter personal data, access
credentials or customer data of the Client into these tools. Confidential
project information shall be processed only in the AI environment provided
by the Client or in a model operated locally by the Contractor. The Contractor
reviews all AI-assisted results before delivery and remains responsible for
them. On request, the Contractor discloses which deliverables were created
with AI assistance. New tools are added to Annex [X] by written agreement.Where this fits
AI clauses are the newest item on a list that keeps growing. NIS2 already pushes security duties down the supply chain, as in NIS2 supplier clauses for freelance contractors, and the status questions around freelancing haven't gone away, see the Statusfeststellung reform. The pattern is the same each time: the clause you can actually comply with is better than the one that looks strict on paper.
If you want a second pair of eyes on an AI clause, request a review.
Related field notes: