What happens legally when AI transforms copyrighted works?

What happens legally when AI transforms copyrighted works?

18 pages · NILP and identity · CIP Legal Working Group · Updated March 2026

Derivative rights, moral rights of integrity, and the NILP Downstream Obligation as the three independent legal exposures created when an AI system generates output that is recognisably derived from a specific creator's work, voice, or likeness.

The three independent exposures

When an AI system generates output recognisably derived from a specific creator's work, three independent legal claims arise:

• Derivative rights: The right to control adaptations and transformations of a copyrighted work. Under UK CDPA s.21, EU Copyright Directive, and US 17 U.S.C. § 106(2), the creation of a derivative work requires the rights holder's authorisation.
• Moral rights of integrity: The right to object to derogatory treatment of a work. Under UK CDPA s.80, this right persists independently of economic rights and cannot be assigned. AI-generated outputs that distort or mutilate the original work engage this right.
• NILP Downstream Obligation: Where the output uses a creator's name, image, likeness, or persona, the CIP framework's NILP Downstream Obligation runs from the AI platform through commercial users to the rights holder — regardless of contractual intermediaries.

Threshold tests

The paper sets out the doctrinal basis for each exposure, the threshold tests applicable to AI-generated material, and the evidentiary requirements for a rights holder asserting any of the three claims.

For derivative rights, the threshold test is whether the AI output incorporates a "substantial part" of the original work (UK) or is "substantially similar" (US). For AI outputs, this analysis applies to both the expressive content and the structural features of the original.

For moral rights, the threshold is whether the treatment of the work amounts to "distortion, mutilation, or other prejudicial treatment" of the work. AI transformations that alter the style, tone, or context of the original while retaining recognisable elements may engage this test.

For NILP, the threshold is recognition — if a reasonable person would identify the output as depicting or impersonating the creator, the NILP Downstream Obligation is engaged regardless of whether the AI operator intended to create a likeness.

Evidentiary requirements

Rights holders asserting any of these three claims need to establish provenance — that the AI system was trained on their work and that the output derives from it. The CIP framework's CDR system and Output-Provenance architecture provide the evidentiary infrastructure for this.