I. Overview

Artificial intelligence systems are often trained on large bodies of text, images, audio, code, and other data. For Philippine authors, the legal question is whether their copyrighted works may have been copied, scraped, stored, transformed, or otherwise used in the development of AI models without permission.

This issue sits at the intersection of copyright law, technology, platform liability, data scraping, contractual licensing, privacy, evidence, jurisdiction, and remedies. In the Philippine context, the core statute is the Intellectual Property Code of the Philippines, particularly the provisions on copyright, economic rights, moral rights, limitations on copyright, infringement, damages, injunctions, and enforcement.

AI training data litigation is still a developing field. Philippine courts have not yet produced a mature body of decisions specifically addressing whether AI model training on copyrighted works is infringement, fair use, or some other legally permissible use. As a result, Philippine authors must often analyze the issue by applying established copyright principles to new technical facts.

This article discusses the legal framework, possible causes of action, evidentiary challenges, jurisdictional issues, defenses, remedies, and practical steps for Philippine authors whose works may have been used in AI training data.

This is general legal information, not legal advice.

II. Who Is a “Philippine Author”?

For purposes of this topic, a Philippine author may include:

a Filipino novelist, poet, essayist, journalist, academic, playwright, blogger, screenwriter, songwriter, software developer, artist, photographer, illustrator, or researcher;
a Philippine resident whose works are protected under Philippine copyright law;
a Philippine corporation, publisher, studio, school, media company, or content platform that owns copyright;
an author whose work was first published in the Philippines;
an author whose work is protected in the Philippines through treaty obligations and domestic law;
an overseas Filipino author whose works are exploited online and may have been included in datasets.

The author is not always the copyright owner. Copyright ownership may belong to a publisher, employer, commissioner, assignee, heirs, or a company depending on contract, employment, inheritance, or assignment.

Before suing, the author must identify whether they personally own the relevant rights.

III. What Works Are Protected?

Philippine copyright law protects original intellectual creations in the literary, scholarly, scientific, and artistic domains. In the AI training context, the most relevant works include:

books;
essays;
articles;
poems;
scripts;
academic papers;
blogs;
news reports;
manuals;
educational materials;
photographs;
illustrations;
comics;
musical compositions;
lyrics;
source code;
databases, if original in selection or arrangement;
translations;
adaptations;
compilations;
audiovisual scripts and subtitles.

Copyright protects the expression of ideas, not the ideas themselves. An author cannot generally claim ownership over a concept, topic, fact, style, genre, method, or historical event. The claim must be tied to protected expression.

For example, an author may own the text of an essay about Philippine history, but not the historical facts discussed in the essay.

IV. Why AI Training Raises Copyright Issues

AI model training may involve several technical steps that are legally significant:

collecting or scraping works from the internet or other sources;
downloading or copying files;
converting works into machine-readable formats;
cleaning, tokenizing, filtering, or annotating data;
storing copies in datasets;
using the works to train a model;
retaining portions of data in embeddings, weights, or intermediate files;
generating outputs that may resemble or reproduce parts of the original works;
commercializing the model or service.

Copyright claims may arise at different stages. The strongest claim may not always be about the final AI output. It may be about the unauthorized copying, storage, or reproduction of works during data collection and training.

V. Copyright Rights Potentially Implicated

A Philippine author may argue that AI training implicates several exclusive economic rights.

1. Reproduction right

The reproduction right is central. If a copyrighted work is copied into a training dataset, downloaded into storage, duplicated in preprocessing, or reproduced in a database, that may be characterized as reproduction.

Even temporary or intermediate copying may matter, depending on the facts and the applicable legal analysis. The author’s argument is that training cannot occur without copying the protected work at least at some stage.

2. Adaptation or transformation right

If the work is converted, extracted, encoded, embedded, summarized, tokenized, translated, or transformed into another form, an author may argue that this implicates the right to make adaptations or derivative works.

AI companies may respond that tokenization or statistical processing is not a derivative work because it does not recast or reproduce the author’s protected expression in a human-readable form. This issue remains highly contested.

3. Distribution right

If datasets containing copyrighted works are shared, sold, licensed, mirrored, or distributed among developers, researchers, contractors, or affiliates, the distribution right may be implicated.

A claim may be stronger where the plaintiff can show that actual copies of the work were included in a dataset made available to others.

4. Public display or communication

For visual works, written materials displayed in datasets, search tools, previews, or outputs may raise public display or communication issues.

For text-based works, public communication may arise if the AI system outputs substantial portions of the author’s text to users.

5. Moral rights

Philippine copyright law recognizes moral rights. These may be relevant if AI outputs reproduce, distort, mutilate, misattribute, or omit attribution from an author’s work in a manner prejudicial to the author’s honor or reputation.

Moral rights claims may include:

failure to attribute authorship;
false attribution;
distortion of a work;
use of the author’s name in connection with degraded or altered content;
modification prejudicial to the author’s reputation.

However, moral rights claims are often fact-specific. The author must show how the use affected attribution, integrity, or reputation.

VI. Is AI Training Itself Copyright Infringement?

This is the central legal question. Under Philippine law, there is no simple settled answer specific to AI training.

A Philippine author’s infringement theory may be:

the AI company copied the copyrighted work without consent;
the copying was not covered by any license;
the copied work was used for commercial AI development;
the work was retained in a dataset or model pipeline;
the use harmed the existing or potential market for the work;
the AI system can generate outputs that compete with or substitute for the work;
the AI company benefited from the author’s labor without compensation.

The AI company’s likely defense may be:

the data was lawfully accessed;
the use was transformative;
the system learned patterns rather than stored expressive works;
the output does not reproduce protected expression;
the copying was temporary or incidental;
the use falls under fair use or a statutory limitation;
the plaintiff cannot prove the work was actually included;
the plaintiff cannot prove substantial similarity or market harm;
the claim should be governed by another jurisdiction’s law.

In Philippine litigation, the outcome would likely depend on facts: what was copied, how much was copied, how the data was used, whether the work was publicly available, whether terms of use prohibited scraping, whether the output reproduces the work, and whether the use harms the author’s market.

VII. Fair Use in the Philippine Context

Philippine copyright law recognizes fair use. Fair use is not automatic. It is determined case by case.

The common factors include:

the purpose and character of the use;
the nature of the copyrighted work;
the amount and substantiality of the portion used;
the effect of the use upon the potential market for or value of the copyrighted work.

1. Purpose and character of use

AI companies may argue that training is transformative because the model does not simply republish the work but uses it to learn statistical relationships in language or media.

Authors may respond that commercial AI training is not a socially neutral research use but an industrial-scale commercial exploitation of protected expression. They may also argue that building a paid AI service from copyrighted books, articles, or art is economically exploitative.

A Philippine court would likely examine whether the use merely replaces the work, creates something meaningfully different, or appropriates the value of the author’s expression.

2. Nature of the copyrighted work

Creative works receive stronger protection than purely factual works. Novels, poems, essays, scripts, songs, illustrations, and photographs may receive stronger protection than factual reports or technical documents, though factual works are still protected in their original expression.

3. Amount and substantiality

Training datasets may include entire works. Authors may argue that wholesale copying weighs against fair use. AI companies may argue that copying entire works is technically necessary for training and that the model does not expose the entire work to users.

The legal significance of full-work copying is one of the most important unresolved issues.

4. Market effect

Authors may argue that AI systems harm their market by:

generating substitute summaries;
imitating their style;
producing competing works;
reducing licensing demand;
replacing commissioned writing;
enabling users to obtain content without buying the original;
weakening the market for authorized AI training licenses.

AI companies may argue that generalized model training does not substitute for the specific work and may even increase discovery of authors. The market-effect analysis may be especially important in Philippine litigation.

VIII. Philippine Authors and Foreign AI Companies

Many AI companies, dataset hosts, cloud providers, publishers, and platforms are located outside the Philippines. This creates jurisdictional issues.

A Philippine author may need to consider:

whether to sue in the Philippines;
whether to join a lawsuit abroad;
whether foreign courts allow class actions or group claims;
whether the AI company has assets or operations in the Philippines;
whether the work was accessed from Philippine servers or websites;
whether the infringement occurred in the Philippines;
whether the defendant’s services are offered to Philippine users;
whether Philippine courts can obtain jurisdiction over the defendant;
whether a foreign judgment can later be enforced in the Philippines.

In practice, many major AI training lawsuits are filed in jurisdictions where defendants are based or where class action mechanisms are stronger. A Philippine author may participate if eligible, but eligibility depends on the rules of that case.

IX. Possible Causes of Action for Philippine Authors

1. Copyright infringement

This is the primary claim. The author must prove ownership of copyright and unauthorized exercise of exclusive rights.

Key issues include:

whether the work is original and protected;
whether the plaintiff owns the copyright;
whether the defendant copied the work;
whether the copying was substantial or actionable;
whether the defendant had authorization;
whether a limitation or defense applies;
whether damages can be proven.

2. Violation of moral rights

This may apply if the AI system misattributes, omits attribution, modifies, distorts, or outputs the work in a way that harms the author’s reputation.

Moral rights are especially relevant for poets, novelists, artists, photographers, journalists, academics, and creators whose reputation depends on attribution and integrity.

3. Breach of contract or terms of use

If the author’s work was hosted on a website with terms prohibiting scraping, automated extraction, commercial reuse, or machine learning training, there may be a contractual theory.

This is often more available to website operators, publishers, or platforms than to individual authors, unless the author controlled the website or was a party to the terms.

4. Unfair competition

An author or publisher may argue that an AI company unfairly profited from their works or created a substitute product. However, unfair competition generally requires specific elements and is not a catch-all remedy for every unauthorized use.

5. Unjust enrichment

The author may argue that the AI company benefited from copyrighted works without payment. This may be pleaded alongside copyright claims, but it may face preemption, duplication, or doctrinal objections depending on the forum.

6. Data privacy claims

If the works contain personal information, or if the author’s personal data was collected in connection with scraping, the Data Privacy Act may become relevant.

For ordinary published works, copyright is usually the primary issue. But if unpublished manuscripts, correspondence, biographies, personal profiles, or identifying information were scraped, privacy claims may arise.

7. Consumer protection or misrepresentation

If an AI service falsely claims that outputs are original, licensed, authorized, or free of infringement, consumer protection or misrepresentation theories may be relevant. These claims would usually be more indirect for individual authors.

X. Ownership Issues: Who Has Standing to Sue?

Before filing any claim, the author must determine who owns the copyright.

1. Individual authors

If the author wrote and published the work independently, they may own the copyright.

2. Employees

If the work was created in the course of employment, ownership may depend on whether the work is part of the employee’s regular duties and on the employment agreement.

For example, a journalist employed by a media company may not personally own all economic rights in articles written as part of employment.

3. Commissioned works

If a work was commissioned, the contract matters. The commissioning party may own certain rights, while the creator may retain others unless assigned.

4. Publishers

Publishing contracts may transfer or license rights. A book author must review the publishing agreement to see whether the author retained digital, electronic, AI training, database, translation, adaptation, or derivative rights.

Older contracts may not mention AI training. This may create disputes between authors and publishers over who can license works for AI training.

5. Co-authors

Co-authored works require analysis of joint ownership. One co-author may not always be able to sue or license without considering the rights of others.

6. Heirs and estates

Copyright may pass to heirs. Heirs of deceased Philippine authors may have rights if the copyright term has not expired. They must establish succession, ownership, and authority to sue.

XI. Evidence Needed in an AI Training Data Claim

Evidence is often the hardest part. It is not enough to suspect that a work was used. The author must gather proof.

Useful evidence may include:

copies of the original work;
publication records;
copyright registration or deposit records, if any;
ISBN, ISSN, DOI, URL, or archive links;
contracts showing ownership;
screenshots of the work online;
proof the work was accessible to scraping;
dataset documentation naming the source;
leaked or public dataset indexes;
search results showing the work in a known dataset;
AI outputs reproducing distinctive passages;
prompts used to generate infringing outputs;
expert analysis comparing output to the original;
evidence of market harm;
evidence of defendant’s commercial use;
takedown correspondence;
admissions in technical papers, documentation, or model cards.

Philippine authors should preserve evidence carefully. Web pages may change. Datasets may be removed. AI outputs may vary over time. Screenshots should include dates, URLs, prompts, and full output context where possible.

XII. Copyright Registration and Proof

Copyright protection generally exists upon creation of the work. Registration is not usually required for copyright to exist.

However, registration or deposit may help prove:

authorship;
date of creation;
ownership;
the content of the work;
priority over later copies;
seriousness of the claim.

For Philippine authors, registration or deposit with the appropriate Philippine copyright office may be useful before litigation, especially when the author expects enforcement problems.

Registration does not guarantee victory, but it helps create an evidentiary record.

XIII. AI Outputs That Reproduce Philippine Works

A claim may be stronger if the AI system outputs verbatim or near-verbatim passages from the author’s work.

The author should test carefully and preserve:

exact prompts;
date and time of generation;
name and version of the AI system;
full output;
comparison with original text;
repeated attempts showing reproducibility;
screenshots or screen recordings;
account settings and location;
any system citations or source references.

A single short phrase may not be enough. Stronger evidence includes substantial similarity, distinctive passages, unique errors, unusual sequence of words, fictional names, invented facts, or other markers showing copying.

XIV. Style Imitation and Philippine Authors

Many authors worry that AI systems can imitate their writing style.

Copyright generally protects expression, not style alone. A claim based only on “the AI writes like me” may be difficult unless the output copies protected expression or specific original elements.

However, style imitation may support other theories if combined with:

use of the author’s name in prompts;
false endorsement;
market substitution;
unfair competition;
violation of moral rights;
reproduction of distinctive protected elements;
contractual restrictions;
publicity or personality rights, where applicable.

For example, “write a new poem in the style of a living Filipino poet” may be ethically troubling but not always copyright infringement by itself. If the output reproduces lines, characters, structure, or distinctive expressive material, the legal claim becomes stronger.

XV. Unpublished Manuscripts and Private Materials

If an unpublished manuscript is used in AI training, the author’s claim may be stronger.

Unpublished works raise additional concerns:

lack of authorization;
breach of confidentiality;
violation of privacy;
breach of submission terms;
unauthorized access;
loss of first-publication value;
moral rights harm;
market damage.

Examples include manuscripts submitted to a publisher, writing contest, thesis repository, editorial platform, cloud storage service, or private workshop. If those materials were used without consent, contractual and privacy issues may arise in addition to copyright.

XVI. Works Posted Online: Does Public Availability Mean Free Use?

No. Posting a work online does not automatically place it in the public domain.

A Philippine author who posts a poem, essay, photograph, or article online generally retains copyright unless they expressly license it away or the term of protection has expired.

However, online posting may affect defenses and factual issues, such as:

whether the defendant had access;
whether the website terms allowed crawling;
whether robots.txt or technical restrictions existed;
whether the author used an open license;
whether the use was expected or implied;
whether the work was copied from a third-party mirror;
whether the author authorized the platform to sublicense content.

Authors should review the terms of any platform where they posted their works. Some platforms obtain broad licenses from users, although those licenses may not always clearly cover AI training.

XVII. Creative Commons and Open Licenses

If a Philippine author released a work under a Creative Commons or other open license, the scope of the license matters.

Important questions include:

Was commercial use allowed?
Were derivatives allowed?
Was attribution required?
Was share-alike required?
Was the license revoked or still valid?
Did the AI company comply with the license conditions?
Was the work obtained from a site that misrepresented the license?

A work under an open license is not necessarily free for all uses. License conditions may still be enforceable.

XVIII. Public Domain Works by Philippine Authors

Some works are no longer protected because the copyright term has expired. These may be in the public domain.

If a Philippine author’s work is in the public domain, copyright claims may no longer be available, although moral rights, attribution norms, cultural heritage rules, or unfair practices may still be discussed depending on the facts.

For living authors and recently deceased authors, copyright protection is more likely still active. For older works, the term must be checked carefully.

XIX. Collective Actions and Class Suits

Individual AI training claims may be expensive. Authors may consider collective action.

Possible collective approaches include:

joining foreign class actions, if eligible;
organizing through authors’ guilds or writers’ unions;
coordinating with publishers;
filing representative actions where legally available;
negotiating collective licenses;
forming a rights management entity;
submitting regulatory complaints;
pursuing test cases.

Philippine procedural rules may not provide the same class action mechanisms as some foreign jurisdictions. However, collective advocacy and coordinated claims may still increase leverage.

XX. Jurisdiction and Choice of Law

AI training disputes may involve multiple countries:

the author is in the Philippines;
the work was posted on a Philippine website;
the server is abroad;
the dataset was compiled abroad;
the AI company is incorporated abroad;
the model was trained abroad;
the service is offered worldwide;
the output is generated in the Philippines.

This creates complex questions:

Which court has jurisdiction?
Which country’s copyright law applies?
Where did infringement occur?
Can the defendant be served with summons?
Can a Philippine judgment be enforced abroad?
Can a foreign judgment be enforced in the Philippines?
Are arbitration clauses or platform terms involved?

For practical enforcement, suing where the defendant has assets or business presence may matter as much as the legal merits.

XXI. Remedies Available to Philippine Authors

A successful copyright claimant may seek several remedies.

1. Injunction

An injunction may seek to stop:

further copying;
continued use of the work in datasets;
distribution of infringing datasets;
generation of infringing outputs;
commercialization of a model trained on unauthorized data;
use of the author’s name or works in specific prompts or services.

In AI cases, injunctions can be technically and commercially complex. Courts may ask whether removal of specific works from a trained model is feasible.

2. Damages

Damages may include actual damages, profits attributable to infringement, statutory damages where available, or other monetary relief depending on the claim and forum.

For authors, damages may be difficult to quantify. Possible theories include:

lost licensing fees;
lost book sales;
lost subscriptions;
lost commissions;
market dilution;
unjust profits;
harm to derivative licensing markets;
harm to future AI licensing opportunities.

3. Accounting of profits

The author may seek profits earned from the infringing use. In AI cases, defendants may argue that profits are attributable to many factors, not any individual work.

4. Destruction or removal

A claimant may seek deletion or removal of infringing copies from datasets, archives, servers, or products. Technical feasibility will be disputed.

5. Attribution or correction

For moral rights violations, remedies may include correction of attribution, removal of false attribution, or measures to address reputational harm.

6. Settlement and licensing

Many disputes may resolve through settlement, licensing fees, opt-out systems, attribution systems, dataset removal, or future royalty arrangements.

XXII. Criminal Liability

Copyright infringement may have criminal aspects under Philippine law in appropriate cases. However, AI training disputes are usually complex commercial disputes and may not easily fit ordinary criminal enforcement unless there is clear, willful, large-scale infringement, piracy, fraud, or unauthorized distribution of copies.

Criminal complaints should be approached carefully. They require stronger factual clarity and may not be suitable for speculative claims.

XXIII. Administrative and Enforcement Avenues

Philippine authors may consider non-court remedies or administrative approaches, such as:

sending demand letters;
filing takedown notices with platforms;
requesting removal from datasets;
notifying publishers or platforms;
approaching the Intellectual Property Office of the Philippines for appropriate procedures;
mediation or alternative dispute resolution;
complaints to data protection authorities if personal data is involved;
coordination with writers’ associations, publishers, or collecting societies.

Administrative and platform remedies may be faster than litigation but may not produce compensation.

XXIV. Demand Letters to AI Companies

A demand letter should be carefully drafted. It may include:

identification of the author;
proof of ownership;
list of works involved;
evidence that the works were used or likely used;
description of infringement;
request for preservation of evidence;
request for disclosure of dataset inclusion;
demand to cease use or remove works;
demand for compensation or licensing negotiation;
reservation of rights;
deadline for response.

The letter should avoid unsupported accusations. A poorly drafted letter may weaken credibility or create defamation risk.

XXV. Preservation of Evidence

Before sending a demand letter or filing suit, the author should preserve evidence.

Recommended steps:

save copies of original works;
save publication dates and metadata;
download webpage archives where lawful;
take screenshots with timestamps;
record prompts and outputs;
preserve account logs;
keep copies of platform terms;
document licensing history;
record sales and revenue data;
preserve correspondence with publishers and platforms;
avoid altering original files.

Evidence preservation is especially important because AI systems change frequently.

XXVI. Proving Dataset Inclusion

One major problem is that authors often cannot directly prove their works were included in training data.

Possible ways to prove inclusion include:

public dataset search tools;
dataset documentation;
URLs listed in training corpora;
leaked dataset indexes;
identical output passages;
memorization tests;
expert statistical analysis;
admissions by developers;
discovery in litigation;
evidence that the defendant scraped the website where the work appeared;
evidence that the work was part of a known book or article corpus.

Speculation is not enough. A viable lawsuit needs evidence or a procedural path to obtain evidence.

XXVII. Discovery Problems

Discovery is the legal process of obtaining evidence from the opposing party. In AI cases, authors may need discovery of:

training datasets;
source URLs;
data vendors;
preprocessing records;
model documentation;
licensing agreements;
output logs;
memorization tests;
data retention policies;
opt-out records;
removal procedures.

AI companies may resist disclosure by invoking trade secrets, confidentiality, security, burden, or irrelevance. Courts may use protective orders to balance secrecy with the author’s need for proof.

In the Philippines, discovery may be more limited than in some foreign jurisdictions. This affects litigation strategy.

XXVIII. Role of Publishers

Publishers may have stronger practical claims than individual authors if they control large catalogs and have clear ownership or licensing rights.

However, authors must examine whether publishers have the right to sue or license AI training uses. Disputes may arise where:

the author retained electronic rights;
the contract predates AI technology;
the publisher licensed works to AI companies without author consent;
royalties for AI licensing were not paid;
the contract does not clearly cover machine learning uses;
the publisher owns only print rights.

Authors should review publishing contracts before assuming who has enforcement authority.

XXIX. Academic Authors and University Works

Philippine academics may face special issues.

Works may include:

journal articles;
theses;
dissertations;
lecture notes;
modules;
textbooks;
research datasets;
conference papers;
course materials.

Ownership may depend on university policy, employment contracts, research grants, journal publishing agreements, and open-access licenses.

Academic authors should check whether they assigned copyright to journals or publishers. They should also check whether open-access publication involved licenses allowing broad reuse.

XXX. Journalists and News Writers

Journalists may have claims if their articles were copied into AI training datasets. However, ownership may belong to the media company if written in the course of employment.

News articles contain both facts and protected expression. The facts themselves are not protected, but the journalist’s original wording, selection, arrangement, and narrative expression may be.

Media companies may also assert claims based on systematic scraping, subscription bypassing, database rights where applicable, unfair competition, or breach of website terms.

XXXI. Software Developers

AI training on source code raises additional issues.

A Filipino software developer may claim infringement if copyrighted code was copied into training data and the AI system outputs substantially similar code.

Important issues include:

whether the code was open source;
what license applied;
whether attribution was required;
whether copyleft obligations were triggered;
whether the generated code is substantially similar;
whether the output includes comments, variable names, or unique structure;
whether the plaintiff owns the code.

Open-source code is not the same as public-domain code. License compliance remains important.

XXXII. Visual Artists and Illustrators

For Filipino visual artists, AI training may involve images scraped from portfolios, social media, online galleries, book covers, comics, or marketplace pages.

Claims may include:

reproduction of images in datasets;
generation of substantially similar images;
style imitation combined with protected elements;
false attribution;
moral rights violations;
commercial substitution;
removal of watermarks or metadata;
use of works in prompt examples or model marketing.

Visual art cases often require expert comparison and careful analysis of protected expression.

XXXIII. Musicians and Lyricists

For Filipino songwriters and lyricists, AI training may involve lyrics, compositions, recordings, or metadata.

Lyrics are literary works and may be protected. AI outputs that reproduce lyrics may be actionable. Musical style imitation alone may be harder to claim unless protected musical expression is copied.

Separate rights may exist in:

lyrics;
melody;
arrangement;
sound recording;
performance;
publishing rights;
neighboring rights.

Ownership may be split among composers, lyricists, publishers, labels, performers, and collecting societies.

XXXIV. Translators and Adaptors

Translations and adaptations may be protected as derivative works if original. A Filipino translator may own copyright in the translation, while the underlying author owns rights in the original work.

AI training claims involving translations require identifying:

who owns the original work;
who owns the translation;
whether permission existed;
whether the translation was copied;
whether the output reproduces the translation’s expression.

XXXV. Government Works and Public Materials

Works of the Philippine government may have special treatment under copyright law. Some government texts may not be protected in the same way as private works, although prior approval may be required for certain exploitation, and private annotations, compilations, translations, or value-added materials may still be protected.

Authors should distinguish between:

official laws, regulations, decisions, and government issuances;
privately authored commentaries;
annotated codes;
legal textbooks;
headnotes;
summaries;
databases;
teaching materials.

A legal author may not own the text of a statute but may own original commentary explaining it.

XXXVI. AI Training and Philippine Data Privacy

Copyright and privacy are different. Copyright protects original expression. Privacy protects personal information and related rights.

AI training may create privacy issues if the dataset includes:

personal essays;
memoirs;
private correspondence;
medical narratives;
student records;
unpublished manuscripts;
personal blogs;
social media posts;
biographical details;
images of identifiable persons.

A Philippine author may have both copyright and privacy concerns if their personal data was processed without lawful basis. Remedies may involve privacy regulators or separate civil claims.

XXXVII. Contractual Protection for Authors Going Forward

Philippine authors should consider adding AI-related clauses to contracts.

Useful clauses may address:

prohibition on AI training without written consent;
reservation of machine learning rights;
separate compensation for AI licensing;
no scraping or dataset inclusion;
no sublicensing to AI companies;
attribution requirements;
audit rights;
data deletion obligations;
warranties against unauthorized AI use;
royalties for AI-related exploitation;
consent for text and data mining;
opt-out mechanisms;
moral rights protection;
disclosure of AI-assisted editing or generation.

Older contracts should be reviewed and updated where possible.

XXXVIII. Licensing AI Training Rights

Some authors may prefer licensing rather than litigation.

An AI training license may define:

covered works;
permitted uses;
model training rights;
whether outputs may compete with the work;
attribution;
compensation;
royalties;
duration;
territory;
exclusivity;
data retention;
deletion rights;
audit rights;
security measures;
restrictions on memorization;
restrictions on verbatim output;
indemnity;
reporting duties.

Authors should avoid granting broad, perpetual, worldwide, sublicensable AI rights without understanding compensation and control consequences.

XXXIX. Opt-Out Systems

Some AI companies or platforms may offer opt-out mechanisms. These may help reduce future use but may not compensate for past use.

Authors should document:

date of opt-out;
works covered;
confirmation receipts;
platform terms;
whether the opt-out applies to training, output, search, indexing, or all AI features;
whether it applies to future models only;
whether third-party datasets are covered.

Opting out should not be treated as a waiver of past claims unless the terms say so. Authors should read terms carefully.

XL. Takedown Notices

If an AI tool, dataset repository, or platform displays or distributes infringing copies, authors may send takedown requests.

A takedown notice should identify:

the copyrighted work;
the infringing material;
location or URL;
proof of ownership;
requested action;
contact information;
statement of good-faith belief;
signature or verification.

Takedowns are more effective for visible copies than for hidden training data.

XLI. Defenses AI Companies May Raise

AI companies may raise several defenses.

1. Fair use

They may argue that training is transformative and does not substitute for the original work.

2. Lack of substantial similarity

They may argue that outputs do not copy protected expression.

3. Lack of proof of copying

They may argue that the plaintiff cannot prove the work was included in training data.

4. License

They may argue that they obtained the work from a licensed provider or under platform terms.

5. Public availability

They may argue that the work was publicly accessible, though public access alone does not defeat copyright.

6. De minimis use

They may argue the use was too minimal to be actionable.

7. Technical necessity

They may argue that intermediate copying is necessary for non-expressive machine analysis.

8. Jurisdictional defenses

They may argue that Philippine courts lack jurisdiction or that another country’s law applies.

9. Safe harbor or intermediary defenses

Platforms and service providers may assert intermediary protections depending on their role.

10. Independent creation

For outputs, they may argue the allegedly similar result was independently generated and not copied from the plaintiff’s work.

XLII. Strengths of a Philippine Author’s Claim

A claim is stronger when:

the work is highly original and creative;
the author clearly owns copyright;
the work was copied in full;
the dataset is known to contain the work;
the AI output reproduces distinctive passages;
the defendant used the work commercially;
no license exists;
the author’s market was harmed;
the work was behind a paywall or subject to anti-scraping terms;
the work was unpublished or confidential;
the AI company ignored takedown or opt-out requests;
multiple works by the same author were included.

XLIII. Weaknesses of a Philippine Author’s Claim

A claim is weaker when:

the author cannot prove dataset inclusion;
the author does not own the copyright;
the work is mostly factual;
only ideas or style were copied;
the output is not substantially similar;
the work was licensed broadly;
the work was released under permissive terms;
damages are speculative;
the defendant is outside Philippine jurisdiction;
the cost of litigation exceeds likely recovery;
the work is in the public domain.

XLIV. Practical Steps for Philippine Authors

A Philippine author concerned about AI training should:

identify the works involved;
confirm copyright ownership;
gather publication and authorship records;
preserve copies and metadata;
search for evidence of dataset inclusion where available;
test AI outputs carefully and preserve results;
review publishing and platform agreements;
register or deposit works where helpful;
send takedown or opt-out requests if appropriate;
consult counsel on demand letters or litigation;
coordinate with other authors or publishers;
consider licensing opportunities;
update future contracts to address AI rights.

XLV. Sample Rights Reservation Clause

Authors may use language similar to the following, subject to legal review:

The Author expressly reserves all rights relating to artificial intelligence, machine learning, text and data mining, dataset creation, model training, model fine-tuning, embedding, automated summarization, synthetic content generation, and similar computational uses. No rights are granted to reproduce, store, scrape, index, process, adapt, tokenize, encode, or otherwise use the Work for the development, training, testing, improvement, or commercialization of artificial intelligence systems without the Author’s prior written consent and separate compensation.

This clause should be customized for publishing, employment, licensing, or platform agreements.

XLVI. Sample Demand Letter Outline

A demand letter may follow this structure:

1. Introduction Identify the author and works.

2. Ownership State the basis of copyright ownership.

3. Infringing use Describe the suspected or confirmed AI training use.

4. Evidence Attach screenshots, dataset references, outputs, URLs, or expert findings.

5. Legal basis Refer to copyright, moral rights, contract, or other applicable claims.

6. Demands Request disclosure, removal, preservation of evidence, compensation, licensing negotiation, and cessation of infringing use.

7. Deadline Set a reasonable response date.

8. Reservation of rights State that all rights and remedies are reserved.

XLVII. Sample Notice to Preserve Evidence

A short preservation demand may state:

Please preserve all documents, datasets, source records, URL lists, training data records, preprocessing logs, model training records, licenses, vendor agreements, output logs, evaluation records, and communications relating to the collection, copying, storage, processing, training, fine-tuning, testing, or deployment of AI systems using the works identified in this letter.

This type of notice is useful before litigation because AI-related evidence may be altered, deleted, or overwritten.

XLVIII. Philippine Policy Considerations

AI training disputes raise broader policy questions for the Philippines:

How should local authors be compensated for machine use of their works?
Should Philippine law create a specific text-and-data-mining exception?
Should authors have an opt-out right?
Should AI companies disclose training data sources?
Should collective licensing be encouraged?
How should Filipino-language works be protected?
How can small authors enforce rights against foreign companies?
Should educational and research uses be treated differently from commercial AI products?
How should moral rights apply to synthetic outputs?
Should government support authors in AI licensing negotiations?

These questions remain open and may require legislation, regulation, industry standards, or test litigation.

XLIX. Special Concern: Filipino-Language and Regional-Language Works

Philippine authors writing in Filipino, Cebuano, Ilocano, Hiligaynon, Waray, Kapampangan, Bikol, Pangasinan, Tausug, Maranao, and other Philippine languages may face unique issues.

Their works may be especially valuable for AI training because Philippine-language datasets are relatively scarce compared with English datasets. This may increase the economic value of local-language works for AI companies.

Potential harms include:

uncompensated extraction of scarce linguistic data;
distortion of regional language usage;
misattribution of cultural works;
generation of low-quality or culturally inaccurate outputs;
exploitation of indigenous or community knowledge;
loss of licensing opportunities for local authors and publishers.

Authors and cultural institutions may wish to assert stronger contractual controls over local-language corpora.

L. Indigenous Cultural Works and Traditional Knowledge

Some works may involve indigenous cultural expressions, oral histories, chants, rituals, designs, symbols, or community knowledge.

These raise issues beyond ordinary copyright. Copyright law may not fully protect communal, traditional, or intergenerational cultural expressions. Other laws and ethical frameworks may apply, especially where indigenous communities’ rights, consent, cultural integrity, and misappropriation are involved.

AI training on indigenous cultural materials should be approached with special care. Consent, attribution, cultural sensitivity, and community control may matter even where conventional copyright claims are difficult.

LI. Litigation Strategy

Before suing, a Philippine author should ask:

Do I own the copyright?
Which works are involved?
Can I prove copying or dataset inclusion?
Is there an infringing output?
Where is the defendant located?
What court has jurisdiction?
What law applies?
What remedy do I want?
Is collective action available?
Are there licensing or settlement options?
Can I afford litigation?
Will discovery be available?
Is the claim better brought by a publisher or association?
Are there limitation periods or urgent deadlines?

A strong case requires both legal theory and technical evidence.

LII. Recommended Documentation File for Authors

Authors should maintain a file containing:

manuscript drafts;
final published copies;
publication dates;
contracts;
copyright registration or deposit records;
royalty statements;
website terms;
screenshots of online publication;
AI output evidence;
correspondence with platforms;
opt-out confirmations;
takedown notices;
proof of market harm;
expert reports;
list of suspected datasets;
legal correspondence.

Good documentation can determine whether a claim is viable.

LIII. Conclusion

Philippine authors may have viable copyright claims if their works were copied, stored, distributed, or used without permission in AI training datasets or if AI systems generate outputs that reproduce protected expression. The strongest claims are likely to involve clear ownership, identifiable works, evidence of copying, substantial reproduction, commercial exploitation, lack of license, and measurable market harm.

The hardest issues are proof and jurisdiction. Many AI systems are developed abroad, training datasets are often opaque, and AI companies may assert fair use, licensing, technical necessity, or lack of substantial similarity. Philippine law provides a framework for copyright and moral rights protection, but AI training litigation remains legally unsettled.

For now, Philippine authors should preserve evidence, review contracts, reserve AI rights, register or deposit important works where useful, monitor AI outputs, coordinate with publishers or author groups, and seek legal advice before sending demands or filing suit. The legal landscape will likely continue to evolve as courts, regulators, authors, publishers, and AI companies confront the value of human-created works in machine learning systems.