Can Copyright Owners Demand Licensing or Compensation for AI Training?

Last materially reviewed: September 6, 2026

Intellectual Property → AI-Generated Works → AI Training & Enforcement

Direct Answer

Copyright owners can offer, negotiate or demand licenses for AI training, but a demand for payment is not automatically enforceable in every case. Whether an AI developer must obtain permission or pay depends on whether protected rights are implicated, whether the use is already licensed, whether an exception such as fair use applies, and what contracts govern the data.

Legal Status

Licensing rights are established; AI compensation models are developing. WIPO has identified consent, transparency and compensation as central questions in the AI copyright ecosystem, but jurisdictions are taking different approaches.

When Compensation Is Most Plausible

• The developer wants a high-quality proprietary dataset.
• The rights holder controls a valuable archive or repertoire.
• The dataset is difficult to substitute.
• Commercial model use is clearly within the requested license.
• A publisher, label, software vendor or creator collective can license rights at scale.
• The developer wants warranties, provenance and reduced litigation risk.

Possible Compensation Models

• One-time dataset access fee.
• Annual or model-version license.
• Per-work or per-token pricing.
• Revenue share.
• Usage-based royalties.
• Research-only or non-commercial licenses.
• Collective licensing through representative bodies where available.

Why AI Developers May Choose to License Even When Law Is Unsettled

Licensing can buy more than permission. It can provide better metadata, cleaner data, provenance records, rights warranties, update feeds, audit rights and commercial certainty. For enterprise AI, those benefits can matter even before courts resolve every fair-use question.

What Rights Holders Should Negotiate

Define the exact works, training purposes, models, territory, duration, output restrictions, attribution, retention, deletion, sublicensing, audit rights and treatment of future model versions. If payment depends on usage, the agreement needs reliable reporting.

Can a Rights Holder Charge After the Work Was Already Scraped?

A rights holder can propose a retroactive license or settlement, but whether payment is legally owed depends on the underlying infringement analysis. A demand letter should distinguish between a commercial offer and a claim that compensation is legally required.

What About Collective Licensing?

Collective models may become important where individual licensing is impractical. WIPO has highlighted the need for scalable rights infrastructure capable of supporting consent and compensation across industries and jurisdictions.

Related Guides

• AI Training Data Licensing
• Fair Use and AI Training
• Dataset Provenance

Separate a Price Proposal From a Legal Entitlement

A copyright owner can propose commercial terms for uses within the rights they control. Whether another party already owes money is a different question. Establish the protected work, relevant acts, permission and applicable exceptions before presenting a demand as an enforceable debt. The IP Code, Sections 177 and 185 (our IP Code explainer) supplies the economic-rights and fair-use framework; it does not provide a universal per-token training tariff.

Define Exactly What the Buyer Receives

Separate access to files from permission to copy them, train or fine-tune models, build a retrieval index, retain datasets or supply downstream users. Specify the covered works and versions. If the seller controls only part of a collection, identify exclusions and third-party components rather than promising blanket clearance.

Under Section 180 of the IP Code, as amended by Republic Act No. 10372, Section 8, copyright may be assigned or licensed in whole or in part. The precise rights and written terms matter. A negotiation should distinguish an assignment of rights from a limited license, and exclusive from nonexclusive permission.

Three Negotiation Examples

A publisher licenses an archive for internal research

Identify the archive release and the authorized project. Address whether commercial deployment requires another agreement, whether contractors may access the files and what happens when the evaluation ends. Check contributor agreements before including freelance photographs or articles. Owning the website does not answer every question about rights in its contents.

A creator considers revenue sharing

Define the revenue base, deductions, reporting period, payment dates and audit process. A percentage is difficult to value if the contract never defines the product whose revenue is shared. Consider how bundled products, affiliates and later model versions will be treated. These are negotiation choices, not statutory royalty rules.

A developer offers payment after disputed collection

Separate permission for future use from settlement of past allegations. Identify the parties, works, time period and claims covered by any release. Do not sign broad language about all works or all future uses without understanding its effect. A payment description should not imply an admission of infringement unless that is actually agreed.

Make the Agreement Verifiable

Ask what use can realistically be measured. Per-work fees need an agreed inventory; usage-based fees need defined events and records; model-version fees need a clear version boundary. Record how disagreements about counts will be resolved. Avoid payment formulas that depend on technical measurements neither party can produce reliably.

Connect the agreement to dataset provenance records so the licensed material and relevant use can be identified later. Address retention and deletion precisely. Deleting source files, removing a retrieval entry and changing trained model weights are different actions; do not promise that one automatically accomplishes all three.

Prepare a Negotiation Brief

1. List the works and rights you can establish.
2. Define permitted uses, excluded uses and downstream access.
3. Select a measurable payment structure and reporting method.
4. Identify warranties, limits, audit rights and termination consequences.
5. Separate future licensing terms from any settlement of past use.

For collective licensing, verify the organization’s actual mandate over the repertoire and use concerned. Do not assume that representation for one category of royalties includes AI training. Use the training-data licensing guide for the broader agreement and the evidence guide when the negotiation follows an alleged violation.

Frequently Asked Questions

Can a creator simply send an invoice to an AI company?

A creator can propose payment, but an invoice alone does not establish that the company legally owes the amount.

Do publishers have stronger negotiating power than individual creators?

Often, because they may control larger catalogues and standardized rights, but ownership and contract chains still need verification.

Can licensing cover future models?

Yes, if the agreement clearly defines future versions, retraining and downstream use.

Official Sources

• IPOPHL — Copyright licensing and transfers
• WIPO — transparency, consent and compensation
• WIPO — Generative AI: Navigating Intellectual Property
• U.S. Copyright Office — Generative AI Training report

Important: This article provides general educational information about Philippine law and technology. It is not legal advice and does not create an attorney-client relationship. Laws, agency procedures, platform terms and the facts of each situation may change the result. Verify current requirements through the cited official sources and seek qualified professional advice when your rights, deadlines, money or legal exposure may be affected.

Featured image: Photo by Vitaly Gariev via Unsplash.