03 · What You Need to Know
Understanding the Legal and Contractual Scope of Publisher AI Restrictions
Publishers May Control Certain Uses of Copyrighted Content
Copyright law grants rights holders specified exclusive rights in qualifying protected works.
These may include rights concerning reproduction and distribution.
Publishers may acquire relevant rights through copyright transfer, licensing agreements, or other arrangements.
However, not every publisher owns every right in every article it distributes.
Researchers should distinguish the publisher's actual legal rights from general claims concerning its content.
The scope of protection also depends on applicable copyright law and exceptions.
Publisher Policies and Binding Contracts Are Not Identical
A publisher may publish guidance describing acceptable uses of its content.
Separately, a researcher or institution may be subject to contractual terms governing access.
These instruments may differ in their legal significance.
For example, a general webpage statement may not fully describe permissions granted through a negotiated institutional license.
Researchers should identify the agreement that actually governs their access.
They should avoid treating every publisher policy statement as equivalent to a binding contractual prohibition.
Subscription Agreements May Restrict Automated Processing
Institutional subscriptions commonly establish conditions for accessing and using scholarly content.
These agreements may address systematic downloading, automated extraction, text and data mining, redistribution, and third-party access.
For example, a license may permit ordinary scholarly reading while requiring additional authorization for large-scale automated processing.
However, the terms vary among publishers and institutions.
Researchers should consult the relevant agreement rather than assume that all subscription journals impose identical conditions.
Publisher Restrictions May Address External AI Services
Some publishers may distinguish local research use from transmission of content to external AI providers.
For example, uploading a subscription-based article to a third-party service may involve additional copying or processing arrangements.
These activities may raise contractual questions concerning authorized users and permitted uses.
Researchers should examine whether the relevant agreement permits the contemplated external processing.
The fact that an AI service accepts the upload does not establish authorization from the content rights holder.
AI Processing Can Involve Several Legally Distinct Activities
The expression AI processing covers a range of technical operations.
These may include document ingestion, text extraction, indexing, summarization, embedding generation, model training, and output reproduction.
Different activities may exercise different rights or fall within different permissions.
For example, authorization to summarize a lawfully accessed article does not necessarily establish permission to redistribute its full text.
Researchers should describe the actual processing workflow rather than treat every use of AI as legally equivalent.
Text and Data Mining May Have Specific Legal Protections
Some jurisdictions recognize copyright exceptions for particular text and data mining activities.
For example, Directive (EU) 2019/790 contains provisions concerning text and data mining for scientific research and other purposes.
These provisions operate subject to specified conditions.
Researchers should examine the relevant jurisdiction, eligible activities, lawful access requirements, and other statutory limitations.
A general publisher prohibition does not necessarily resolve whether a statutory exception applies.
European Research Mining Exceptions Have Particular Conditions
Article 3 of Directive (EU) 2019/790 addresses certain reproductions and extractions by research organizations and cultural heritage institutions for scientific research.
Article 4 provides a separate framework for text and data mining, subject to conditions including the treatment of rights reservations.
These provisions should not be treated as identical.
Their implementation and application also depend on relevant national law.
Researchers should avoid assuming that all AI processing qualifies automatically as protected text and data mining.
Rights Reservations Can Matter for Some Mining Activities
Under the European Union framework, particular text and data mining activities may be affected by rights reservations.
For example, Article 4 of Directive 2019/790 addresses situations in which rights have been expressly reserved in an appropriate manner.
However, the operation of such reservations should be distinguished from the separate scientific research exception under Article 3.
Researchers should examine the specific statutory provision rather than assume that every reservation has the same legal effect.
Contractual Restrictions May Not Override Every Statutory Exception
Some copyright laws limit the ability of contractual provisions to prevent activities authorized by particular statutory exceptions.
For example, Article 7 of Directive (EU) 2019/790 addresses the enforceability of contractual provisions contrary to specified exceptions.
However, these protections do not necessarily apply to every AI use or jurisdiction.
Researchers should distinguish a publisher's contractual conditions from the legal effect of applicable statutory rights.
Consequential disputes require jurisdiction-specific interpretation.
United States Fair Use Requires Contextual Analysis
In the United States, fair use may permit certain uses of copyrighted material without separate permission.
However, its application depends on the statutory factors and circumstances of the use.
Academic or noncommercial purpose may be relevant but is not automatically determinative.
Researchers should avoid treating fair use as a universal authorization for uploading publisher content to any AI system.
They should also distinguish fair-use analysis from contractual obligations.
Philippine Copyright Law Has Its Own Framework
In the Philippines, Republic Act No. 8293, the Intellectual Property Code, as amended, governs copyright and relevant limitations.
Researchers should examine the applicable provisions when evaluating publisher restrictions on computational use.
Foreign legal doctrines may offer comparative background but do not automatically determine the Philippine position.
The enforceability of a particular contractual restriction may depend on the actual agreement and applicable law.
Qualified legal guidance may be appropriate where substantial rights are involved.
Open-Access Licenses Can Grant Broad Reuse Permissions
Some publishers distribute articles under Creative Commons or other open licenses.
These licenses may grant permissions for copying, adaptation, and redistribution subject to specified conditions.
For example, CC BY generally permits broad reuse, including commercial reuse, subject to attribution and other applicable requirements.
Researchers should examine the license attached to the actual article.
A publisher's general AI policy should not automatically be assumed to revoke rights already granted under an applicable license.
Publishers Cannot Necessarily Revoke Existing Open Licenses
Creative Commons licenses are generally designed to be irrevocable while the licensee complies with their terms.
Consequently, a publisher's later policy change does not ordinarily revoke permissions validly granted under an existing Creative Commons license.
However, the researcher must still comply with the license and consider any separately protected third-party material.
Researchers should verify the actual license and its conditions rather than rely solely on a general publisher policy.
Open Access Does Not Mean Every Activity Is Permitted
Open-access publications may be subject to different licenses.
For example, noncommercial or no-derivatives conditions may affect particular forms of reuse.
Researchers should determine whether the contemplated AI activity exercises rights covered by those conditions.
They should also distinguish private computational analysis from public redistribution of adapted material.
Open access provides important permissions, but those permissions should be interpreted accurately.
Third-Party Material May Be Excluded From an Article's License
Research articles may include figures, photographs, or other material owned by separate rights holders.
Such content may not be covered by the publisher's general open-access license.
Researchers should inspect relevant credit lines and licensing notices.
For example, an AI workflow that extracts and republishes a third-party illustration may require additional permission.
Publisher AI Policies May Distinguish Research and Commercial Uses
Some licensing arrangements differentiate between academic research and commercial exploitation.
However, the classification of a particular AI workflow may be more complicated than whether the researcher personally receives payment.
For example, processing may involve a commercial AI provider even when the research project is noncommercial.
The legal significance depends on the license wording and applicable law.
Researchers should avoid assuming that every activity involving a commercial service is automatically commercial use under every license.
Model Training May Raise Different Issues From Summarization
Using copyrighted publications to train or fine-tune an AI model may involve different technical and legal considerations from summarizing a document for an individual researcher.
For example, training may involve repeated processing and retention of substantial collections of works.
Researchers should determine whether the relevant authorization covers the actual activity.
Permission to read or summarize a publication does not automatically establish permission to incorporate it into a model training dataset.
Retrieval-Augmented Generation Requires Source Governance
Retrieval-augmented generation systems may index publications and retrieve relevant passages when responding to questions.
These systems can involve storage, extraction, and reproduction of source material.
Researchers should examine whether the content may be indexed and used in the intended environment.
They should also consider whether generated responses reproduce protected passages beyond applicable permissions.
Authorization for internal retrieval does not necessarily establish unrestricted public redistribution.
Publisher-Provided AI Features May Operate Under Different Terms
Some publishers integrate AI-assisted discovery or summarization into their platforms.
These services may operate under permissions established by the publisher or relevant rights holders.
However, researchers should not assume that content processed within such a platform can be freely exported to unrelated AI services.
The relevant terms should be examined for the specific activity.
Institutional Agreements May Provide Additional Permissions
Universities may negotiate licensing arrangements that permit particular computational research activities.
For example, an agreement may authorize defined text and data mining workflows.
Researchers should consult institutional library or legal personnel when the applicable permissions are unclear.
General publisher website statements may not fully describe negotiated institutional rights.
Different Article Versions May Have Different Rights
A study may be available as a preprint, accepted manuscript, and final published article.
Each version may be distributed under different conditions.
For example, an author may retain permission to share an accepted manuscript through a repository while the publisher's formatted version remains subject to a subscription agreement.
Researchers should identify the rights associated with the actual version they intend to process.
Authors May Retain Certain Rights
Publishers do not necessarily hold every right in every published article.
Authors may retain copyright, grant a license to publish, or reserve specified reuse rights.
Institutional and funding agreements may also influence the arrangement.
Researchers should avoid assuming that a publisher's general policy establishes exclusive ownership of all content it distributes.
Publisher Restrictions Can Affect Research Reproducibility
Restrictions on sharing or processing source documents may complicate reproduction of AI-assisted research workflows.
For example, a researcher may be unable to redistribute a licensed corpus used for computational analysis.
However, reproducibility does not always require unrestricted publication of copyrighted source material.
Researchers may document search procedures, identifiers, code, analytical parameters, and authorized access routes.
The appropriate archival strategy should respect the rights associated with the underlying content.
Restrictions Can Affect Research Equity
Institutions differ in their access to scholarly subscriptions and negotiated computational permissions.
These differences may influence which researchers can conduct particular AI-assisted analyses.
However, the consequences depend on the available open literature, alternative sources, and applicable legal exceptions.
Researchers should evaluate actual access disparities rather than assume that every publisher restriction has identical effects.
Copyright Restrictions Should Not Be Confused With Scientific Quality
A publication's licensing conditions do not independently establish its methodological rigor or relevance.
For example, a subscription-based article may be scientifically valuable, while an openly licensed article may still require critical appraisal.
Researchers should distinguish access and processing permissions from evaluation of the underlying evidence.
Publisher Policies May Change Over Time
Publishers may revise AI processing policies, licensing arrangements, and platform conditions.
Researchers should verify the terms applicable when the processing occurs.
They should also distinguish newly adopted policies from rights granted under earlier agreements.
Changes should be evaluated according to the relevant contractual and legal framework.
Research Institutions Should Provide Clear Guidance
Universities can help researchers interpret scholarly content licenses and identify approved processing arrangements.
For example, library services may maintain guidance concerning publisher-specific text and data mining permissions.
Research governance and information security personnel may also advise on external AI services.
Such guidance can reduce uncertainty without replacing qualified legal interpretation when necessary.
Researchers Should Document the Basis for AI Processing
When AI materially contributes to research through analysis of published content, researchers should retain relevant permission and workflow records.
These may include publication licenses, institutional agreements, processing purposes, and service conditions.
Researchers should also document how generated findings were verified against original sources.
Transparency supports accountability but does not independently establish lawful processing.
Publisher Restrictions Should Be Evaluated, Not Automatically Accepted or Ignored
Researchers should avoid two opposite assumptions.
The first is that publishers can prohibit every computational use of their content regardless of applicable law.
The second is that academic purpose or lawful access automatically overrides all restrictions.
Both positions are too broad.
The appropriate analysis identifies the rights holder, governing agreement, statutory exceptions, and actual processing activity.
Where uncertainty remains, researchers should seek clarification before undertaking consequential uses.
