TDM and Training Data Law

TDM and Training Data Law

This specialist module provides in-depth analysis of text and data mining (TDM) law as it applies to AI training. It covers the three major jurisdictions and their interaction through contracts and enforcement.

UK: Data (Use and Access) Act 2025

The UK DUA Act 2025 replaced the earlier proposed TDM exception (which would have allowed commercial TDM without consent) with a balanced regime:

• A TDM exception exists for non-commercial research purposes (carried over from CDPA 1988 s.29A).
• Commercial TDM (including AI training) is subject to a rights-holder opt-out mechanism.
• The opt-out must be expressed in a machine-readable form accessible to the AI operator.
• Content carrying a valid opt-out signal is excluded from the exception — any use requires explicit licence.

The CIP framework's cip.md declaration satisfies the machine-readable requirement when deployed at the domain root with CIP-Training-Ingestion: Prohibited. The Act does not prescribe a specific format, but the Government's technical guidance references "standardised machine-readable signals" — which cip.md implements.

EU: AI Act Article 53 and Copyright Directive

The EU framework operates through two complementary instruments:

• Copyright Directive Article 4: A TDM exception for commercial purposes, subject to a rights-reservation mechanism. Rights holders who "appropriately" reserve their rights are excluded from the exception.
• AI Act Article 53: Imposes transparency obligations on general-purpose AI model providers, including a requirement to "put in place a policy to comply with Union copyright law, and in particular to identify and comply with reservations of rights expressed pursuant to Article 4(3) of Directive (EU) 2019/790".

The combined effect: AI operators must actively check for and respect rights reservations. The "appropriate" reservation for online content must be machine-readable. The CIP cip.md format satisfies this requirement.

US: Fair use four-factor analysis

The US has no statutory TDM exception or opt-out mechanism. The primary legal framework is the fair use doctrine under 17 U.S.C. § 107, which requires analysis of four factors:

1. Purpose and character of the use — is the AI training use "transformative"? Courts have reached different conclusions on this.
2. Nature of the copyrighted work — creative works receive stronger protection than factual works.
3. Amount and substantiality of the portion used — AI training typically ingests entire works, which weighs against fair use.
4. Effect on the market for the original — does AI training (and AI-generated outputs) substitute for the original? This is the most contested factor.

The NYT v. OpenAI/Microsoft litigation (1:23-cv-11195, S.D.N.Y.) is the most significant active case. The April 2025 motion-to-dismiss ruling allowed most claims to proceed, and the case is now in expert-discovery. No final ruling on fair use in AI training has been issued by a US federal court.

Jurisdiction selection in contracts

The jurisdiction clause in a content licensing contract determines which TDM regime applies. This is a strategic drafting decision:

• UK jurisdiction: Gives the rights holder access to the DUA Act 2025 opt-out mechanism and statutory remedies.
• EU jurisdiction: Gives access to the AI Act Article 53 compliance framework and Copyright Directive rights-reservation regime.
• US jurisdiction: Relies on fair use arguments (uncertain) but provides access to federal copyright statutory damages under 17 U.S.C. § 504(c).

The CIP framework recommends that UK and EU rights holders include a UK or EU jurisdiction clause in content licensing agreements, giving them access to the statutory opt-out mechanisms. US rights holders may prefer US jurisdiction for the statutory damages regime despite the fair use uncertainty.

Enforcement pathways

• UK: Statutory damages under DUA Act 2025 for breach of opt-out; additional damages under CDPA 1988 s.97(2); injunctive relief through the Intellectual Property Enterprise Court or High Court.
• EU: Injunctive relief through national courts implementing the Copyright Directive; Article 53 enforcement through national AI supervisory authorities; GDPR enforcement where personal data is involved.
• US: Statutory damages under 17 U.S.C. § 504(c) ($750–$30,000 per work, up to $150,000 for wilful infringement); actual damages and profits; injunctive relief through federal district courts.

Collecting society coordination

Collecting societies can coordinate Training Data Dividend claims at scale through their existing mandates. The CIP framework supports this through:

• CDR records that identify collecting society membership (PRS, MCPS, DACS, etc.)
• Revenue waterfall routing that can direct Training Data Dividend payments through existing collecting society infrastructure
• Collective licensing negotiations that societies can conduct on behalf of their members using CDR-documented rights exposure data

For rights holders not represented by a collecting society, the Rights Registry provides direct enforcement support and Training Data Dividend distribution.