How to legally prevent agencies from using my work to train AI models

To legally prevent agencies from using your work to train AI models, you must include an explicit restrictive covenant in your service agreement that prohibits the use of your deliverables for machine learning, model training, or algorithmic development. Standard 'work-for-hire' clauses are insufficient; you need specific, affirmative language that carves out AI training rights.

The Legal Landscape of AI Training Rights

In the absence of specific contractual language, agencies often rely on broad 'Work Made for Hire' or 'Perpetual License' clauses to claim rights to your output. Under current US Copyright law, while the ownership of the final deliverable may transfer to the client, the right to use that deliverable as training data for a Large Language Model (LLM) is a distinct commercial right that should be negotiated separately.

Why Standard Clauses Fail

• Work-for-Hire: Transfers ownership but does not explicitly define the scope of 'use' in the context of AI.
• Broad License Grants: Phrases like 'for any purpose' or 'all commercial uses' are often interpreted by courts to include machine learning.
• Data Processing Clauses: Agencies often hide AI training rights inside standard 'data processing' or 'service improvement' sections.

Key takeaway: If your contract says the agency can use your work for 'any purpose' or 'internal business operations,' you have likely already granted them the right to train AI models on your intellectual property.

Action Item: Review your current Master Services Agreement (MSA) for the phrase 'for any purpose.' If found, strike it and replace it with 'for the specific purpose of delivering the services defined in this SOW.'

Drafting Enforceable AI Restrictions

To effectively block AI training, your contract must contain a specific 'Prohibition on Machine Learning' clause. This clause should be narrow enough to be accepted by the agency but broad enough to protect your creative output.

Essential Clauses to Include

1. Prohibition Clause: 'Client shall not use, or permit any third party to use, the Deliverables or any underlying data provided by the Contractor to train, fine-tune, or develop any artificial intelligence, machine learning, or generative models.'
2. Data Ownership Retention: 'Contractor retains all rights to the raw data, prompts, and intermediate work products used to generate the Deliverables.'
3. Indemnification for AI Misuse: 'Client shall indemnify and hold harmless the Contractor from any claims arising from the unauthorized use of the Deliverables in AI training sets.'

Action Item: Copy the 'Prohibition Clause' above and append it to the 'Intellectual Property' section of your next contract draft.

Jurisdictional Considerations

The legal enforceability of these clauses varies by jurisdiction. In the United States, the Copyright Office has yet to issue definitive guidance on whether training AI on copyrighted works constitutes 'fair use.' However, contract law remains supreme. If you sign a contract that explicitly forbids AI training, the agency is in breach of contract if they do so, regardless of whether the act itself is copyright infringement.

Key Jurisdictional Factors

• California (CCPA/CPRA): Provides stronger protections for personal data, which may overlap with your work output.
• EU (GDPR/AI Act): The EU AI Act introduces transparency requirements that may force agencies to disclose if they are using your work for training.
• New York/Delaware: Courts in these states strictly interpret contract language; be precise with your definitions of 'AI' and 'Machine Learning.'

Key takeaway: Do not rely on copyright law alone. Your contract is your primary shield. If the contract is silent on AI, you are at a significant disadvantage.

Action Item: If you work with international agencies, ensure your contract includes a 'Choice of Law' clause that favors your home jurisdiction to ensure easier enforcement.

How TermScore Protects Your IP

Manually reviewing contracts for hidden AI training clauses is time-consuming and prone to human error. TermScore automates this process by scanning your agreements for high-risk language, identifying clauses that grant agencies broad rights to your work, and suggesting specific, legally-vetted revisions to protect your intellectual property. By using TermScore, you ensure that your creative output remains yours, not fuel for an agency's AI model.