03 · What You Need to Know
Copyright, Licensing, and Data Governance When Uploading Research Papers to AI
Lawful Access Does Not Automatically Authorize Every Use
Researchers may obtain journal articles through institutional subscriptions, personal purchases, open-access platforms, or other lawful channels.
However, the right to access and read a publication is not necessarily identical to the right to reproduce, redistribute, or process it through an external service.
For example, an institutional subscription may permit personal scholarly reading and downloading while imposing additional conditions on automated processing or third-party access.
Researchers should examine the permissions associated with the specific publication and access arrangement.
The broader distinction between access and permitted AI processing is explored in whether lawful access permits uploading research papers to AI services.
Uploading a PDF May Involve Reproduction and Transmission
When a researcher uploads a PDF to a cloud-based AI service, the file may be transmitted to and processed within infrastructure operated by the provider or its subprocessors.
Depending on the system, copies may be created temporarily or retained for specified periods.
These activities can have legal and contractual implications distinct from reading the publication locally.
Researchers should determine whether the relevant license or exception permits the contemplated processing.
The technical details matter because not every AI system handles uploaded files in the same way.
Institutional Library Subscriptions May Contain Usage Restrictions
Universities often provide access to scholarly publications through agreements negotiated with publishers or content providers.
These agreements may define authorized users, permitted uses, automated processing conditions, and restrictions on redistribution.
Some arrangements may permit particular research-related computational activities, while others may require additional authorization.
Researchers should not assume that all institutional subscriptions contain identical terms.
University librarians or electronic resources personnel may be able to clarify the relevant agreement.
Publisher Policies May Restrict Certain AI Processing Activities
Publishers may establish contractual conditions governing automated analysis, text and data mining, or the use of their content with external AI systems.
However, restrictions vary among publishers, journals, licensing arrangements, and jurisdictions.
Researchers should distinguish publisher preferences or contractual conditions from statutory rights and exceptions.
A publisher's general website statement may not fully describe the permissions granted through a particular institutional agreement.
The issue of publisher restrictions on AI processing of scholarly content requires examination of the actual applicable terms.
Open Access Does Not Always Mean Unrestricted AI Processing
Open-access publications may be distributed under licenses that permit various forms of reuse.
However, the precise permissions depend on the license and any excluded third-party material.
For example, an article published under a Creative Commons Attribution license generally permits copying and adaptation subject to its conditions.
Other licenses may impose restrictions concerning commercial use or distribution of adaptations.
Researchers should inspect the license attached to the actual article rather than rely solely on an open-access label.
Creative Commons Licenses Should Be Read Carefully
Creative Commons licenses differ in the rights they grant and the conditions they impose.
For example, CC BY generally permits reuse and adaptation, including commercial reuse, subject to attribution and other applicable conditions.
CC BY-NC includes a noncommercial limitation, while licenses containing ND restrict sharing adapted material under specified circumstances.
Whether a particular AI workflow constitutes an adaptation, reproduction, or another legally relevant use depends on the activity and applicable law.
Researchers should avoid assuming that every restriction applies identically to private analysis and public redistribution.
Article Licenses May Exclude Third-Party Figures
A research paper may contain photographs, illustrations, diagrams, or other material licensed separately from the article itself.
For example, a figure credit may indicate that an image is reproduced with permission from another rights holder.
The article's general open-access license may not necessarily cover that image.
Researchers should examine figure credits and licensing notices when the AI workflow involves extracting, reproducing, or redistributing such material.
The permissions associated with the article and its individual components may differ.
Personal Purchase Does Not Necessarily Grant Unlimited Reuse Rights
Researchers may purchase access to an individual article from a publisher.
However, purchasing access does not necessarily transfer copyright or authorize unrestricted redistribution.
The transaction may grant a license to access or use the publication under specified conditions.
Researchers should review those terms before transmitting the file to an external AI provider.
Ownership of a downloaded copy should not be confused with ownership of the intellectual property rights in its contents.
Public-Domain Publications May Offer Broader Reuse Opportunities
Some research publications or other scholarly materials may be in the public domain.
For example, copyright protection may have expired or may not apply to particular works under the relevant law.
Such material may be reusable without copyright permission.
However, public-domain status can differ across jurisdictions, and particular editions or accompanying materials may have separate rights.
Researchers should verify the status of the actual document before relying on it.
Copyright Exceptions May Permit Certain Research Uses
Copyright systems may provide exceptions or limitations relevant to research, education, quotation, or computational analysis.
However, their scope differs across jurisdictions.
For example, some legal systems contain specific text and data mining exceptions, while others apply broader doctrines such as fair use.
Researchers should evaluate the applicable provision and its conditions rather than assume that every academic AI upload is exempt.
The legal treatment may depend on whether the activity involves lawful access, commercial processing, retained copies, or subsequent redistribution.
Fair Use Is Not a Universal Permission
In the United States, fair use involves a contextual assessment of statutory factors.
Academic or research purposes may be relevant, but they do not automatically determine the outcome.
Other countries may apply different legal tests.
Researchers should not treat a general explanation of United States fair use as sufficient authorization for uploading publications in every jurisdiction.
Where legal uncertainty has substantial consequences, qualified advice may be necessary.
Philippine Copyright Law Requires Specific Consideration
In the Philippines, Republic Act No. 8293, the Intellectual Property Code, as amended, provides the principal statutory framework for copyright.
Relevant provisions concerning economic rights, limitations, and fair use may affect particular research activities.
However, the permissibility of uploading a copyrighted article to a specific AI service cannot be determined from the academic purpose alone.
Researchers should examine applicable Philippine law alongside relevant publisher and institutional agreements.
Foreign copyright guidance should not automatically be treated as controlling local law.
Text and Data Mining Is Not Identical to Conversational AI Uploading
Text and data mining generally involves computational analysis of text or other information to identify patterns or extract information.
Uploading a PDF to a generative AI assistant may involve additional activities, including document storage, retrieval, summarization, and generated reproduction of source material.
The distinction matters because licenses or statutory exceptions may define permitted activities in particular ways.
Researchers should examine whether the actual AI workflow falls within the relevant authorization.
Permission for one type of computational analysis should not automatically be assumed to cover every AI service or processing arrangement.
AI Service Terms Can Affect Document Handling
AI providers may establish different policies concerning uploaded files, data retention, model improvement, subprocessors, and organizational controls.
These arrangements may differ between consumer and institutional services.
Researchers should review the terms applicable to the specific account and service.
They should also determine whether uploaded documents can be deleted, exported, or accessed by other parties under the applicable arrangements.
Provider policies should be evaluated using current documentation rather than assumptions about how all AI tools operate.
Uploading to a Third-Party Service May Be Contractually Significant
Institutional licenses sometimes distinguish authorized users from external organizations or services.
Uploading licensed content to an external AI platform may therefore require additional consideration.
Researchers should establish whether the processing arrangement is permitted by the relevant agreement.
The mere fact that the researcher remains the person operating the AI interface does not necessarily resolve contractual restrictions concerning third-party processing.
Local Processing May Present Different Considerations
Researchers may use locally operated software or models rather than transmitting documents to an external cloud service.
Local processing can reduce certain disclosure and data transfer concerns.
However, it does not automatically eliminate copyright restrictions on copying or computational analysis.
Researchers should examine the applicable permissions regardless of whether processing occurs locally or remotely.
The technical architecture affects the risk assessment but does not independently establish lawful use.
AI Summarization Does Not Necessarily Require Full-Text Uploading
Researchers may sometimes obtain useful assistance by providing limited bibliographic information, their own notes, or appropriately selected excerpts.
For example, AI can help organize a research matrix based on verified information extracted by the researcher.
However, whether excerpts may be transmitted still depends on applicable rights and circumstances.
Using less material can reduce unnecessary disclosure, but it should not be treated as a universal legal exemption.
Researcher-Prepared Notes Can Be an Alternative
Researchers may independently read articles and prepare notes describing research questions, methods, findings, and limitations.
AI can then assist with organizing those notes or identifying comparisons.
However, the notes should not reproduce extensive protected expression without appropriate authorization.
The researcher should also preserve accurate attribution to the original publications.
This approach may reduce dependence on uploading complete copyrighted documents while maintaining direct engagement with the evidence.
AI-Generated Summaries Still Require Verification
Even when uploading an article is permitted, the resulting AI summary may be inaccurate.
Generative AI can omit limitations, misstate sample characteristics, or attribute findings incorrectly.
Researchers should compare consequential summaries with the original publication.
Permission to process a document does not establish that the generated interpretation is scientifically reliable.
AI May Reproduce Protected Passages in Its Response
An AI assistant may quote or closely paraphrase material from an uploaded article.
The resulting output can raise questions distinct from the legality of the initial upload.
For example, publishing a substantial passage from an article may require permission even when private analysis of the document was authorized.
Researchers should distinguish internal processing from public reproduction or redistribution.
The possibility of AI reproducing published research articles should be considered when reviewing generated outputs.
Confidential Manuscripts Require Additional Protection
Researchers may receive unpublished manuscripts through peer review, collaboration, or institutional responsibilities.
Such documents may be subject to confidentiality obligations independently of copyright.
For example, a peer reviewer should not assume that uploading a confidential manuscript to an external AI service is permitted merely because the document was provided for review.
Researchers should follow the relevant journal, funder, institutional, or contractual confidentiality requirements.
Where authorization is absent, the material should not be disclosed to an external service.
Peer Review Material Should Be Treated With Particular Care
Peer review frequently involves access to unpublished research under conditions of confidentiality.
Some publishers expressly restrict the use of generative AI with confidential submitted manuscripts.
Researchers should review the applicable journal policy before using AI during peer review.
The ability to upload a manuscript does not establish permission to do so.
Confidentiality and research integrity obligations remain relevant even when the intended AI task is only summarization.
Research Papers May Contain Personal or Sensitive Information
Some research documents include identifiable participant information, unpublished datasets, or confidential appendices.
Uploading these materials may raise data protection and research ethics concerns.
Researchers should examine the content of the actual document before transmission.
Where personal data are involved, applicable consent, lawful processing grounds, institutional policies, and security arrangements may matter.
Copyright permission does not replace data protection compliance.
Data Retention and Model Improvement Policies Matter
Researchers should determine whether the AI service retains uploaded documents and whether submitted content may be used for model improvement.
These conditions can vary by provider, product, account type, and configuration.
For example, an institutional arrangement may offer different data handling protections from an individual service.
Researchers should verify the actual policy rather than assume that every uploaded PDF is either permanently stored or immediately deleted.
Institutional Approval May Be Necessary
Universities may establish policies governing the use of external AI tools with licensed publications or research materials.
These policies may involve library services, information security, research governance, or legal offices.
Researchers should confirm whether the proposed workflow requires institutional authorization.
Such requirements should be distinguished from copyright law itself.
Internal approval does not necessarily override third-party contractual restrictions.
Publisher Agreements May Differ From General Website Policies
A publisher's publicly available AI policy may not fully describe the terms of an institutional subscription.
For example, a university may have negotiated specific permissions concerning text and data mining.
Researchers should consult the agreement governing their actual access.
Where uncertainty remains, the library or responsible institutional office may be able to request clarification.
Publisher Restrictions Should Be Evaluated Against Applicable Law
Contractual restrictions and statutory copyright exceptions are not identical.
Whether a contract can restrict a particular statutory exception depends on the jurisdiction and relevant legal framework.
Researchers should avoid assuming that publisher terms always override every legal exception.
They should also avoid assuming that a research exception automatically invalidates contractual obligations.
Consequential conflicts require appropriate legal analysis.
Uploading One Article Is Different From Bulk Processing
Processing a single article for a particular research task may involve different technical and contractual considerations from uploading thousands of subscription-based publications.
Bulk processing may implicate automated access restrictions, text and data mining conditions, or infrastructure limitations.
Researchers should examine the scale and manner of processing.
Permission for individual scholarly reading should not automatically be interpreted as authorization for large-scale extraction or redistribution.
Downloading Through a University Library Does Not Transfer Copyright
Institutional subscriptions generally provide access under contractual arrangements.
They do not ordinarily transfer ownership of the copyright in every downloaded article to the researcher.
Researchers should therefore avoid treating possession of a PDF as equivalent to unrestricted control over its contents.
The rights associated with the publication remain governed by applicable law and agreements.
AI Tools With Integrated Scholarly Access May Have Different Permissions
Some research platforms integrate AI functions with scholarly content through arrangements involving publishers, repositories, or licensed databases.
Such integrations may provide permissions that differ from uploading the same material to an unrelated external service.
However, researchers should not assume that every integrated platform has unrestricted rights to all scholarly content.
The relevant authorization depends on the actual agreements and features.
Research Collaboration Does Not Automatically Permit Document Sharing
Researchers may collaborate across institutions with different library subscriptions.
One collaborator's lawful access does not necessarily authorize sharing a licensed PDF with every partner or uploading it to a shared AI workspace.
Researchers should verify the relevant sharing permissions.
Where appropriate, collaborators can use lawful access routes, open versions, or independently prepared bibliographic information.
Preprints May Have Different Licensing Conditions
Preprints are often available through public repositories, but their reuse conditions vary.
Some are distributed under explicit open licenses, while others retain more restrictive rights.
Researchers should inspect the license or terms attached to the particular preprint.
Public availability does not automatically establish unrestricted permission for external AI processing.
Accepted Manuscripts and Published Versions May Have Different Rights
Researchers may encounter preprints, accepted manuscripts, and final publisher versions of the same study.
These versions may be governed by different licensing and distribution arrangements.
For example, an author may be permitted to share an accepted manuscript through a repository while the published version remains subject to a subscription agreement.
Researchers should verify the rights associated with the actual version they intend to process.
Researchers Should Record the Basis for Permitted Processing
Where AI-assisted document processing forms a material part of a research workflow, researchers should retain relevant documentation.
This may include the publication license, institutional authorization, AI service terms, and the purpose of processing.
Records may also identify the service used, date of processing, and relevant data handling conditions.
Such documentation can support transparency and institutional accountability.
However, retaining records does not independently establish that an unauthorized activity was lawful.
Copyright Compliance Does Not Establish Research Integrity
A researcher may lawfully upload an article while receiving an inaccurate AI-generated summary.
Conversely, a scientifically accurate summary may have been produced through a workflow that violates contractual restrictions.
These concerns require separate evaluation.
Researchers should verify both the permissions governing document processing and the accuracy of the resulting research content.
The Safest Workflow Depends on the Actual Permissions
No single rule can determine whether every copyrighted research paper may be uploaded to every AI service.
The appropriate decision depends on the publication's rights, applicable legal framework, institutional agreements, service architecture, and intended processing.
Researchers should verify these conditions before uploading protected material.
Where permission remains uncertain, they may use authorized alternatives, seek clarification, or work from independently prepared notes.
AI assistance should support research without bypassing the rights and responsibilities associated with scholarly content.
