Most creative contracts are built for a world where AI didn’t exist. They cover scope, payment, revisions, and ownership, and they do it reasonably well. What they don’t do is say anything about what happens when AI is part of the work. According to Envato’s 2026 State of AI in Creative Work report (1,780 creators surveyed), 58% of creative professionals are using AI tools on client projects. Only 28% of agency owners disclose this to clients. That gap between usage and transparency is now more than a professional courtesy issue. The EU AI Act’s Article 50 transparency obligations take full effect on August 2, 2026, less than three weeks from the date of this article. (If you want to know exactly what Article 50 requires of your creative team, we’ve put together a compliance checklist for creative teams before August 2.)
This article gives you three plain-language clauses you can add to your existing contracts today, plus guidance on how to raise them without unsettling a client relationship.
Legal disclaimer: This article is for educational purposes only. It is not legal advice. The clause language below is illustrative: it has not been reviewed by a legal practitioner for your specific jurisdiction or engagement type. Consult a qualified IP attorney or contracts solicitor before modifying your contracts.
Three gaps most creative contracts leave open
The AI use gap. Your contract almost certainly doesn’t say whether you can use AI tools on a client project at all. This creates a problem in both directions: if a client later objects to AI involvement they didn’t know about, your contract offers no protection. And if a client wants to restrict your AI use in ways that would break your workflow, there’s nothing to negotiate against.
The IP gap. Standard assignment clauses transfer copyright from creator to client. That logic assumes the creator holds the copyright to transfer. The US Copyright Office’s January 2025 guidance requires “sufficient human control” over expressive elements for copyright to subsist. If AI generated substantial portions of a deliverable, the assignment may not transfer what either party thinks it does. This is not a theoretical risk. It’s an unresolved gap sitting in the middle of most client handovers right now. For a fuller picture of what the recent AI copyright ruling means for your practice, see [what the AI copyright ruling means for your creative work](/insights/the-future-of-work/ai-copyright-ruling-creative-work).
The reverse gap. This one rarely appears in contracts, and no standard template addresses it. Once you hand over a deliverable, what can the client do with it? Feed it into their internal AI model? Generate variants from it? Strip your authorship from it entirely? Your standard contract almost certainly says nothing about this. The clause you need runs in both directions.
The three clauses to add
These are three short, standalone additions to your existing contract language. Each one addresses a distinct risk.
Clause 1: AI use and disclosure
This clause establishes that you’ll notify the client before using AI tools to generate or substantially draft a deliverable, and that use requires their written approval. It protects you against retroactive objections and protects the client’s right to know how their work was produced.
Most clients accept this clause without friction. Presenting it signals that you take both quality and transparency seriously.
Illustrative language (adapt for your jurisdiction):
> Supplier will notify Client in writing before using any AI system, large language model, image generation tool, or similar automated tool (“AI Tools”) to generate, substantially draft, or substantially modify any Deliverable. Use of AI Tools for these purposes requires Client’s prior written approval. Supplier will identify the specific tools used and describe how they were applied.
Clause 2: IP assignment with human authorship warranty
The risk here is specific: if you’ve used AI to generate substantial portions of a deliverable, you may not hold the copyright you’re purporting to assign. This clause doesn’t restrict your use of AI. It requires you to warrant that the deliverable contains sufficient human creative contribution for copyright to subsist, and that you’ve disclosed any AI-generated material before the assignment takes effect.
This protects both parties. The client gets a warranty they can rely on. You get a clear standard to hold yourself to before signing off.
Illustrative language (adapt for your jurisdiction):
> Supplier assigns to Client all right, title, and interest in the Deliverables. Supplier warrants that: (a) the Deliverables contain sufficient human creative contribution such that copyright subsists in the Deliverables under applicable law; (b) Supplier is the author of the human-authored elements; and (c) Supplier has not incorporated AI-generated material in any Deliverable without prior disclosure and Client approval.
Clause 3: Training data restriction
This clause runs in both directions. It stops client data from being fed into AI tools’ training pipelines without consent. It also stops clients from taking your finished deliverables and using them to train, fine-tune, or prompt their own AI models after handover. That second direction is the one almost no one thinks to include.
Multiple analyses of the EC’s draft guidelines conclude that B2B contracts can address training data restrictions explicitly, and that doing so offers clearer recourse than relying on default IP provisions alone.
Illustrative language (adapt for your jurisdiction):
> Neither party will use, or permit their tools or service providers to use, the other party’s data, materials, or deliverables to train, fine-tune, or improve any AI or machine learning model without the other party’s prior written consent. This obligation extends to all third-party AI tools used in the course of this engagement.
None of these clauses are adversarial. They document how you actually work. Any client who uses AI tools themselves will recognize the language immediately and understand exactly what it’s doing.
How these apply across disciplines
- Designer: Clause 1 protects you when Midjourney or Firefly is part of your comp or ideation process and the client hasn’t explicitly approved AI in the brief.
- Developer: Clause 2 matters when GitHub Copilot suggests substantial portions of production code you’re about to assign as original work.
- Copywriter: Clause 3 stops a client from feeding your final copy into their own model to generate variants, derivatives, or campaign iterations without your knowledge or consent.
How to raise these without killing a deal
With a new client, include the clauses in your standard contract template alongside scope and payment terms. The framing is simple: “Here’s how I work with AI tools. This protects both of us.” You’re not asking for special treatment. You’re setting up a professional baseline. (Deciding whether the engagement is worth taking on at all? Our guide to how to vet clients before you say yes covers the signals worth checking before contracts are even on the table.)
With an existing client, a short email works well. Propose a contract amendment and reference the August 2026 EU AI Act date as the practical trigger. Most clients are aware the regulation exists even if they don’t know the details. Framing it as an update you’re making across all engagements takes the personal pressure out of it.
When a client sends you their own AI policy clause, read it carefully with Clause 3 in mind. Does it restrict your AI use so broadly it would break your workflow? Does it claim the right to use your deliverables for AI training? Those are the red flags. A client’s AI policy is not automatically balanced just because their legal team drafted it.
The contracts most creators are using were written before any of this existed. Adding three short clauses doesn’t change the nature of the work. It just makes the terms honest about how the work actually gets done.
The Blue Mango is a creative co-op where fair contracts and transparent AI workflows are the baseline, not the exception.