Last materially reviewed: September 6, 2026
Intellectual Property → AI-Generated Works → AI Training & Enforcement
Direct Answer
Fair use may be relevant to AI training in the Philippines, but AI training is not automatically fair use. Section 185 of the Intellectual Property Code requires a case-by-case analysis of four familiar factors: purpose and character of the use, nature of the copyrighted work, amount and substantiality used, and effect on the potential market or value of the work.
There is not yet a Philippine Supreme Court decision that definitively resolves generative-AI training under these factors. That makes careful factual analysis essential.
Legal Status
Developing. The fair-use statute is settled law; its application to modern generative-AI training remains unsettled. IPOPHL’s 2026 discussions have specifically placed fair use and data scraping within the current AI-copyright debate.
The Four Fair-Use Factors
1. Purpose and character
Courts consider why the work was used and whether the use is commercial or non-profit educational. A developer may argue that model training analyzes works to learn statistical relationships rather than distribute the originals. A rights holder may respond that commercial model development competes in markets built on copyrighted expression. Neither characterization is automatically controlling.
2. Nature of the copyrighted work
Highly creative works such as novels, songs, illustrations and photographs generally present stronger copyright interests than bare factual material. Datasets often mix both.
3. Amount and substantiality
Training may involve complete copies of works. Using the whole work does not automatically defeat fair use, but it is an important factor. The analysis can also consider whether the amount used was reasonably necessary for the asserted purpose.
4. Market effect
This can become the hardest question. Rights holders may point to existing or emerging licensing markets for AI training. Developers may argue that training does not substitute for access to the original works. The answer can vary by industry and dataset.
Why Foreign AI Cases Are Not Philippine Law
Courts and copyright offices abroad are actively addressing generative-AI training, licensing and fair-use questions. Those developments can be useful comparative material, but a U.S., EU or other foreign ruling does not automatically decide a Philippine dispute.
The U.S. Copyright Office’s 2025 Part 3 report on generative-AI training is therefore best treated as comparative analysis, not a statement of Philippine law.
Decision Snapshot
- Non-profit research dataset: purpose may weigh differently from a commercial foundation model, but all factors still matter.
- Creative works copied in full: nature and amount factors require careful analysis.
- Licensed dataset available: market-effect arguments may become more significant.
- Model outputs reproduce source works: output-side infringement can create a separate issue from training.
For Businesses Building or Fine-Tuning AI
Do not rely on the phrase “fair use” as a compliance shortcut. Document why each dataset is needed, what rights attach to it, whether licenses are available, what restrictions apply, how much of each work is used, and how the model is tested against memorization or regurgitation.
See AI Training Data Licensing in the Philippines and Dataset Provenance and AI Copyright.
A Four-Factor Assessment Worksheet
Section 185 is the controlling starting point. It lists purposes including research but still requires examination of the statutory factors. The worksheet below organizes the relevant facts; it is not a points system or an official AI-training approval test. IP Code, Section 185.
| Factor | Questions to document | Useful evidence |
|---|---|---|
| Purpose and character | What does the project actually do? Is the use commercial, educational or mixed? | Research plan, product specification and deployment plan |
| Nature of the work | Does the source contain factual reporting, creative expression, or both? | Representative source works and content classifications |
| Amount and substantiality | Which portions are copied, and why? Are qualitatively important portions included? | Corpus inventory, extraction rules and retained copies |
| Market effect | How might the use affect demand or value? What evidence supports either position? | Licensing records, intended outputs and market analysis |
Include facts unfavorable to the proposed use. A memo that says only “research is fair use” does not evaluate the other factors. Likewise, a rights holder should identify the affected work and market instead of assuming that all commercial computation necessarily produces the same result.
Four Shortcuts to Avoid
No fixed percentage. Section 185 does not give AI developers a universal permitted fraction. Amount includes substantiality, so a small but important part can matter. No automatic clearance from attribution. Giving credit does not replace permission or an exception. No decisive project label. An internal or educational label must match the actual use. No automatic answer from output tests. Training-stage reproduction and later outputs require separate analysis. IP Code, Sections 177 and 185.
Distinguish the law from arguments about its application. A developer may argue that training serves a different purpose from reading a source work. That is an argument to evaluate under Philippine law, not a Philippine court holding that all model training is lawful. Foreign decisions must be identified by jurisdiction and actual scope before being used comparatively.
Three Hypothetical Assessments
A Limited Academic Study
A researcher proposes a small corpus for a defined experiment. Document the scope, access arrangements, copies retained and dissemination. The research purpose is relevant, but does not eliminate the need to consider the work, amount and market. Reassess if the project later becomes a commercial service.
A Commercial Model Using Complete Books
A developer copies full creative works for a paid product. Record why complete copies are used, the nature of the corpus, licensing availability and expected uses. Do not make a categorical conclusion from commerciality alone, but do not ignore it. Obtain qualified review before relying on an exception for a high-value or disputed corpus.
A Model Produces Long Matching Passages
Outputs reproduce distinctive portions of source works. Preserve the outputs, prompts, model version and source comparisons. This raises a separate issue about use of protected expression, even where the training team previously prepared a fair-use assessment. See AI output infringement.
What the Assessment File Should Contain
Keep the dataset identity, proposed acts, source rights, factor-by-factor reasons, unresolved facts, reviewer and approved scope. Preserve any licenses that cover part of the corpus; a mixed dataset may need different treatment for different sources. Do not present an unsupported confidence score as legal clearance.
Define when review must reopen: new data, new commercial purpose, redistribution, a credible complaint, or a material change in output behavior. These are practical governance triggers, not statutory deadlines. The organization should know which dataset and purpose the earlier assessment actually covered.
If the Case for Fair Use Is Unclear
- Identify the unresolved factor or missing fact.
- Reduce or exclude disputed material while it is assessed.
- Investigate whether a suitable license is available.
- Use material with clearer rights where practical.
- Obtain qualified Philippine legal advice for the actual project.
- Document the decision and any limits before proceeding.
Licensing is one way to resolve uncertainty, not proof that every unlicensed use infringes. Conversely, a pending licensing discussion does not authorize current use. See AI training licenses and the training-material decision guide for the alternative bases and limits.
Frequently Asked Questions
Is research always fair use?
No. Research is one of the purposes named in Section 185, but the four-factor analysis still applies.
Does commercial use automatically defeat fair use?
No. Commerciality matters, but it is one factor among several.
If the AI never shows the original work, is training automatically lawful?
No. Training-stage copying and output-stage reproduction are related but distinct questions.
Official Sources
- Republic Act No. 8293 — Section 185 Fair Use (our IP Code explainer)
- IPOPHL — Copyright
- IPOPHL — 2026 copyright forum on fair use and data scraping
- U.S. Copyright Office — AI study
- WIPO — Artificial Intelligence and Intellectual Property
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. Fair use is fact-specific.
Featured image: Photo by Sasun Bughdaryan via Unsplash.