ChatGPT Library Sharing Governance Guide: Viewer and Editor Roles, Folder Ownership, Access Revocation, and File Custody

Why the September 9 Library sharing update changes governance, not just collaboration

OpenAI’s September 9, 2026 ChatGPT release notes introduce a material change for teams that treat ChatGPT Library as a working knowledge surface: Library users can share files and folders with named people as Viewer or Editor, share items across a workspace, review or change access, remove access, and use shared items from Shared with me in conversations. That is more than a convenience feature. It creates a custody model in which files can move into shared operational use while ownership, uploader attribution, access rights, retention behavior, and deletion obligations remain distinct governance questions.

The most important custody rule in OpenAI’s release note is narrow and consequential: files uploaded into a shared folder belong to the folder owner, while the person who uploaded each file is still recorded. If a contributor uploads a file to another person’s shared folder and later loses access to that folder, the file remains in the owner’s folder and the contributor loses access. Administrators, research leads, legal teams, and department owners should treat that as a records-management event, not simply as a collaboration action.

This guide uses the term custody to describe practical control over where a file resides and who is responsible for managing it inside ChatGPT Library. Custody is not the same as authorship, attribution, retention, or deletion. A product manager may upload a vendor comparison into a finance-owned shared folder, a finance lead may own the folder, the uploader identity may remain recorded, the workspace retention policy may govern the file, and a later access removal may affect only who can reach the file—not whether the file exists.

This article explains enterprise security checks for deploying ChatGPT Work, focusing on data governance, access control, and audit compliance for autonomous workflows across apps, files, websites, and desktops. The complete 3 Enterprise Security Checks Before Deploying ChatGPT Work — Data Governance, Access Control, and Audit Compliance article provides the destination-specific detail for this section’s ChatGPT Workspace Security decision because it is the strongest match for a workspace-level security marker because the current guide’s sharing roles, ownership, revocation, and file custody decisions depend on the same access-control and audit-compliance foundations.

The new sharing primitives: named people, workspace sharing, roles, and Shared with me

OpenAI describes two sharing scopes in the September 9 release note: sharing with named people and sharing across a workspace. Named-person sharing is useful when a file or folder should be limited to a specific group of collaborators, such as a deal team, litigation group, customer-success pod, or research review committee. Workspace sharing is broader and should be treated as a higher-risk governance action because it can make a resource available across a larger organizational boundary, subject to the workspace controls OpenAI references.

The release note identifies two roles for Library sharing: Viewer and Editor. A governance program should not rely on the role names alone; it should define internal decision rules for when each role is appropriate. A Viewer assignment is the safer default for reference material, final reports, approved templates, and read-only research inputs. An Editor assignment is better reserved for people who are expected to contribute to, maintain, or update the shared Library item or folder as part of an accountable workflow.

Shared with me is the operational bridge between access control and ChatGPT use. OpenAI states that users can use shared items from Shared with me in conversations. That means a shared file is not merely visible as a document; it can become context for analysis, drafting, research synthesis, policy comparison, code explanation, or other ChatGPT work. A file-sharing decision therefore affects both human access and model-assisted use inside the account or workspace context.

Teams should be careful not to add capabilities OpenAI has not described. The release note does not say Library sharing is public-link sharing, does not say uploader ownership overrides folder ownership, and does not say removing access deletes files. A practical governance policy should explicitly prohibit users from treating Library folder sharing like consumer document-link sharing unless OpenAI separately documents such behavior for the relevant plan and workspace.

Access, custody, attribution, retention, and deletion are separate controls

The common mistake is to collapse five governance questions into one: “Who has access?” OpenAI’s documentation requires a more precise model. Access determines who can reach a file or folder. Custody determines where the file resides and who owns the folder. Attribution records who uploaded a file. Retention determines how long files persist under the applicable account or workspace policy. Deletion determines whether the file is removed from Library and scheduled for permanent deletion under OpenAI’s documented conditions and exceptions.

The separation matters most during offboarding, incident response, deal-room cleanup, research archive review, and regulated-records handling. For example, if a departing consultant uploaded files into a company-owned shared Library folder, removing that consultant’s access does not remove the files from the owner’s folder under OpenAI’s release note. The correct process is to revoke access, inventory uploaded files, decide whether the folder owner should retain them, and then apply the relevant Library deletion or retention workflow if removal is required.

A practical mental model for Viewer and Editor assignments

Recommendation: treat Viewer as the default for materials that are authoritative, sensitive, final, or used as reference context in conversations. Examples include approved policy documents, board-ready metrics exports, final customer interview summaries, product requirement baselines, legal research packets, and vendor security questionnaires. Viewer access reduces accidental changes to shared materials and gives teams a clearer path for source-controlled use in ChatGPT conversations.

Recommendation: reserve Editor for active collaboration folders where the recipient is expected to add or maintain files. Examples include a research intake folder where analysts upload source PDFs, a sales enablement folder where field leaders contribute updated objections, or an implementation folder where a project manager collects approved customer-provided materials. Because OpenAI states that files uploaded into a shared folder belong to the folder owner, Editor access should be paired with an explicit folder-owner responsibility to review incoming files.

Operational warning: do not grant Editor access merely because someone is senior, cross-functional, or likely to ask for changes. If a user only needs to use a file from Shared with me in a conversation, Viewer may be sufficient depending on the organization’s policy and the actual task. If a user needs to upload files into another person’s shared folder, the uploader should understand that the file remains in the owner’s folder even if the uploader later loses access.

This guide covers shared context across ChatGPT and Codex, including memory, projects, and cross-platform workflows for unified AI work. The complete The Complete Guide to ChatGPT and Codex Shared Context: Memory, Projects, and Cross-Platform Workflows article provides the destination-specific detail for this section’s Shared Project Governance decision because it adds useful context for shared project governance by connecting library-sharing rules to broader project, memory, and shared-context workflows rather than to unrelated project-management content.

Folder ownership creates file custody outcomes that access revocation does not undo

The folder-owner rule is the center of this guide because it affects how teams should design shared Library folders. A folder should have an accountable owner before it becomes a collection point for other people’s files. The owner may be an individual user in the product sense, but the organization should map that ownership to a real operating responsibility: department steward, project lead, records owner, data steward, legal matter owner, or workspace administrator process.

Consider a research team building a competitive-intelligence folder. The head of strategy owns the shared folder, analysts have Editor access, and executives have Viewer access. An analyst uploads interview notes and market PDFs into that folder. Under OpenAI’s documented rule, those files belong to the folder owner, while the analyst remains recorded as uploader. If the analyst rotates off the project and loses folder access, the files remain available to the folder owner and permitted collaborators rather than reverting to the analyst or disappearing automatically.

That outcome is useful when a company needs continuity, but it is risky when users treat uploads casually. A contributor should not upload personally held, externally licensed, customer-confidential, privileged, export-controlled, or contract-restricted material into someone else’s shared folder unless the organization’s policy allows that custody transfer. The governance question is not only “Can the contributor upload?” but also “Should the folder owner become custodian of this file after the contributor leaves?”

Retention and deletion require separate Library procedures

OpenAI’s chat and file retention article states that chats remain in an account until manually deleted, that deleting a chat removes it from the account immediately and schedules permanent deletion within 30 days subject to de-identification, disassociation, or security/legal exceptions, and that archiving hides a chat but does not delete it. For Library governance, the critical rule is different: files uploaded in conversations are saved to Library when Library is available, and chats and Library files are managed separately.

That means deleting a conversation is not a Library cleanup procedure. OpenAI explicitly states that deleting a chat does not delete files saved to Library and that users must view and delete saved files from Library. A user who attaches a spreadsheet to a conversation, later deletes the chat, and assumes the spreadsheet is gone may leave the file active in Library. In a shared-folder setting, the risk expands because the file may also be available to other users through a shared folder or Shared with me.

For Enterprise, Edu, and Healthcare contexts, OpenAI states that Library files follow the workspace retention policy. That makes workspace retention a policy dependency for Library governance. Teams should avoid promising a universal retention period to employees, customers, auditors, or counterparties unless they have verified the applicable workspace policy and any exceptions that may apply. OpenAI’s documentation also notes that legal, security, de-identification, disassociation, backup, expiration, and unsupported or transient-file behavior can affect outcomes in specific cases.

Deep Research makes shared Library hygiene more important

OpenAI’s Deep Research documentation adds another reason to govern Library sharing carefully: Deep Research can use the public web, uploaded files, and supported connected apps when available and authorized, but starting Deep Research does not grant access to additional files or apps. Workspace, role, session-tool, provider, and connected-account permissions continue to apply. In practice, shared Library files can become part of the evidence base for research tasks when the user has access and includes or uses them in the task flow.

Shared source material should therefore be curated before it becomes research context. A stale pricing sheet, outdated legal memo, draft customer transcript, or unapproved product roadmap placed in a shared folder can influence downstream summaries, comparisons, or decision memos. OpenAI also warns in its Deep Research guidance that citations improve traceability but do not guarantee correctness, so a cited report generated from shared Library material still requires human source review before decisions, external publication, or high-stakes use.

The practical governance position is straightforward: Library sharing should be designed around least necessary access, explicit folder custody, accountable upload practices, reviewable attribution, workspace-aware retention, and separate deletion procedures. Treat Shared with me as a work surface that can feed ChatGPT conversations, not as a passive file inbox. Treat folder ownership as custody, not decoration. Treat revocation as an access event, not a deletion event. Treat every shared research source as something that may shape future outputs and must be reviewed accordingly.

Designing an ownership and permissions model for shared Library folders

OpenAI’s September 9 release notes make the custody rule explicit: when a person uploads a file into another person’s shared Library folder, the file belongs to the folder owner, while the uploader’s identity remains recorded. This means governance cannot treat “who uploaded it” and “who controls the file” as the same field. In operational terms, the folder owner is the custodian of the item, the uploader is the attributed contributor, and access revocation changes visibility and use rights without automatically deleting the file.

The most important consequence is offboarding and role-change behavior. If a contributor adds a document to a shared folder and later loses access to that folder, OpenAI states that the file remains in the owner’s folder and the contributor loses access. A team should therefore avoid using shared folders as informal personal drop boxes unless the owner is prepared to retain, review, classify, and eventually delete or preserve the uploaded files under the relevant workspace policy.

A workable governance model should separate five questions before any folder is shared: who owns the folder, who may view it, who may add or modify contents, who is accountable for classification and cleanup, and which retention or legal process applies after a file is no longer actively used. Treating these as separate questions prevents a common mistake: removing a person’s Editor role and assuming that all files they contributed have been removed, transferred, or made inaccessible to every downstream workflow.

Permission matrix for Viewer and Editor assignments

The release notes identify two sharing roles, Viewer and Editor, for files and folders shared with named people, and they also describe workspace-wide sharing and access review or removal. The sources do not define every interface-level operation for each role, so the safest administrative matrix should describe governance intent rather than inventing product behavior. Use the matrix below as a policy layer that your administrators and folder owners can map onto the actual controls available in your workspace.

The matrix should be embedded in the team’s access request process. A request for Editor access should require a reason that cannot be satisfied by Viewer access, a defined collaboration window, and a named approver. A request for workspace-wide sharing should require a classification check because workspace-wide visibility is a distribution decision, not merely a convenience setting.

This Codex Library integration guide explains how teams connect Library assets to existing ChatGPT workflows, providing a practical companion for understanding how shared files, folder scope, permissions, and downstream use need to be governed together. The complete How to Integrate Codex Library with Your Existing ChatGPT Workflows: Complete Migration and Setup Guide article provides the destination-specific detail for this section’s Enterprise Knowledge Access Controls decision because the target directly concerns Codex Library integration and enterprise knowledge flow, making it substantially closer to the marker than a speculative memory-system article.

Least-privilege patterns for shared Library collaboration

Least privilege in Library sharing means giving each person the smallest role that supports their current task and then reviewing the assignment when the task ends. For a research team, analysts who only need to cite an approved briefing should be Viewers, while the two people maintaining the source packet may be Editors. For a sales enablement folder, field teams may be Viewers of finalized battlecards, while product marketing staff retain Editor access to update the canonical versions.

A strong pattern is to create separate folders for intake, working drafts, and approved reference material. The intake folder should have tightly controlled Editor access because contributors can add files that become part of the owner’s folder custody. The working folder should be limited to the active project team. The approved folder can have broader Viewer access once the owner has verified that the material is properly named, classified, and free of content that should not be distributed across the selected audience.

Another useful pattern is time-boxed edit access. Grant Editor rights for a defined contribution period, then switch the person to Viewer or remove access after the owner has accepted, rejected, or moved the submitted files according to the folder’s rules. This prevents dormant Editors from accumulating across long-running folders and makes it easier to explain why a person still has access during an audit.

Naming rules that preserve custody, purpose, and sensitivity

File and folder names should carry enough governance context to survive search, review, and reuse in conversations. A practical naming convention should include the business domain, purpose, sensitivity level, date or version, and owner or steward code where your organization uses one. The goal is not cosmetic consistency; it is to help users avoid attaching the wrong file to a conversation, using obsolete material in Deep Research, or sharing a folder more broadly than the contents justify.

Recommended folder naming pattern:
[Domain] - [Purpose] - [Sensitivity] - [Owner/Steward] - [Lifecycle]

Examples:
Finance - Q4 Board Prep Sources - Restricted - FP&A Steward - Active
Product - Launch Messaging Approved - Internal - PMM Steward - Published
Legal - Vendor Diligence Intake - Confidential - Legal Ops - Review

The classification term should describe the handling requirement, not the perceived importance of the work. For example, “Confidential” should be reserved for information that requires a restricted audience under company policy, while “Internal” may be suitable for approved material intended for ordinary employee use. If the organization already has a data classification standard, Library names should reuse that vocabulary instead of creating ChatGPT-specific labels that conflict with security, records, or legal processes.

Version labels should distinguish drafts from approved references. A file named “Customer interviews final” is weak because it does not identify whether the file is final for analysis, final for legal review, or final for external publication. A stronger label is “Customer Research Synthesis – Approved for Internal Planning – 2026-09-10,” assuming the classification and approval are accurate. Folder owners should reject uploads with ambiguous names when ambiguity could cause improper reuse.

Classification rules before granting folder access

Before a folder is shared, the owner should classify both the folder and its expected contents. A folder that may contain confidential source contracts should not inherit the same access pattern as a folder containing approved training notes. Classification should happen before adding Editors because Editors can contribute files that remain in the owner’s folder after their access ends.

Classification should also account for downstream use in conversations. OpenAI notes that shared items can be used from “Shared with me” in conversations. If a file is shared too broadly, the risk is not only that more people can see it; the risk is that more people may use it as context for new work, summaries, research tasks, or decision support. Access design should assume that shared knowledge may become operational input, not just static reference material.

Owner responsibilities for folders that accept uploads

The folder owner should act as the records steward for files uploaded into the shared folder. That responsibility includes setting the purpose of the folder, deciding who receives Viewer or Editor access, reviewing uploaded files for fit, resolving naming or classification errors, removing access when collaboration ends, and coordinating deletion or retention actions through the appropriate Library and workspace procedures.

Owners should maintain a short folder charter for any shared folder that accepts Editor contributions. The charter should state what the folder is for, what files may be uploaded, what files are prohibited, which classification applies, who may approve new Editors, how often access is reviewed, and what happens when the project closes. A charter can be a simple document, but it should be accessible to contributors before they upload files so they understand that contribution is not the same as ownership.

Sample folder charter fields:
Folder owner:
Business purpose:
Permitted file types or content categories:
Prohibited content:
Classification:
Approved Viewer groups or named users:
Approved Editor criteria:
Contributor acceptance requirement:
Review cadence:
Project close or retention action:
Escalation contact:

Owners should also validate that deletion expectations are realistic. OpenAI’s retention documentation states that chats and Library files are managed separately: deleting a chat does not delete files saved to Library, and users must view and delete saved files from Library. In Enterprise, Edu, and Healthcare contexts, Library files follow the workspace retention policy. Therefore, an owner cannot rely on chat cleanup, access removal, or archiving as a substitute for Library file governance.

Contributor acceptance rules before upload

Contributor acceptance is the policy checkpoint that prevents surprise custody transfers. Before receiving Editor access to an upload-capable folder, contributors should acknowledge that files they upload into the shared folder belong to the folder owner, that their uploader identity may remain recorded, and that losing access later means they lose access to the folder while the file remains with the owner. This acceptance should be especially explicit for contractors, cross-functional contributors, and employees contributing files from another team’s system of record.

A practical acceptance statement should avoid legal overreach and focus on operational facts. It should say that contributors may upload only files they are authorized to share for the folder’s stated purpose, that they must follow naming and classification rules, that they must not use the folder for personal storage, and that they should notify the owner if they upload a file by mistake. The owner should define the remediation path for mistaken uploads because access revocation alone does not delete the file.

Sample contributor acceptance: “I understand that files I upload into this shared Library folder are kept in the folder owner’s Library custody, while my uploader identity remains associated with the upload. I will upload only material I am authorized to contribute for the stated purpose, use the required naming and classification rules, and notify the folder owner promptly if I upload the wrong file or content that should not be retained in this folder.”

Contributor acceptance should be captured before Editor access is granted, not after a problem occurs. For sensitive folders, owners should require contributors to identify the source of each upload, the reason it belongs in the folder, and whether any restrictions apply to reuse. This is especially important where later research or summarization work could combine several uploaded files into a new deliverable.

Periodic access reviews that separate access from file cleanup

Access reviews should ask two different questions: who should still have access, and what files should still remain in the folder. The first question can be answered by checking Viewer and Editor assignments against current roles and project needs. The second requires content review, classification review, and retention awareness. Removing an Editor may be correct, but it does not remove the files the person previously uploaded.

The review record should include the date, reviewer, folder owner, access changes made, unresolved questions, and any file cleanup actions assigned. Avoid vague outcomes such as “reviewed access” because they do not show whether Editors were checked separately from Viewers or whether uploaded files were evaluated after contributor removal.

Decision rules for revocation, transfer, and cleanup

Use revocation when a person no longer needs access to the folder; use cleanup when a file no longer belongs in the folder; use ownership or stewardship reassignment when the accountable owner is no longer the right custodian. These are different operations. Revoking a contributor’s access addresses future visibility and use, but OpenAI’s documented custody behavior means it does not remove the contributor’s uploaded files from the owner’s folder.

1. Revoke access when a contributor leaves the project, changes role, violates folder rules, or no longer has a business need.
2. Review contributed files after revocation to decide whether the files are still valid, correctly classified, and appropriate for the folder’s purpose.
3. Delete through Library procedures only when the owner and applicable policy allow deletion; do not rely on chat deletion or archive actions to remove Library files.
4. Preserve under policy when workspace retention, legal, security, or compliance requirements require continued retention.
5. Reassign stewardship when the folder owner is no longer the right accountable custodian for the business process.

The safest operational warning is simple: access revocation is not a deletion workflow, uploader attribution is not ownership, and chat deletion is not Library cleanup. A mature Library governance program should make those distinctions visible in folder charters, access request forms, contributor acceptance text, and periodic review evidence.

Retention, offboarding, and file custody: what must survive access changes

Library sharing governance fails when teams treat access removal as the same thing as data disposal. OpenAI’s retention guidance separates several lifecycle events that administrators and folder owners must manage independently: deleting a chat, deleting a Library file, archiving a chat, deleting a project or custom GPT, applying workspace retention policy, losing access to a shared folder, and compliance export availability. A leaver process that only removes a person from shared folders may protect future access, but it does not automatically delete files they uploaded into someone else’s folder, does not remove Library files saved from chats, and does not rewrite folder ownership.

OpenAI states that chats remain in an account until manually deleted, and deleting a chat removes it from the account immediately while scheduling permanent deletion within 30 days, subject to de-identification or disassociation and security or legal exceptions. That timing should be treated as a deletion workflow for the chat object, not as a general erasure rule for every file that appeared in the conversation. If Library is available, files uploaded in conversations are saved to Library, and OpenAI’s guidance says chats and Library files are managed separately.

Archiving is a different control from deletion. OpenAI says archiving hides a chat but does not delete it, which means archiving should never be used as an offboarding, legal-hold, data-minimization, or incident-containment substitute. A practical governance rule is simple: archive for personal workspace organization, delete for a deletion request when deletion is permitted, and use workspace retention or legal/security controls for regulated preservation or exception handling.

Library-file deletion requires a separate Library procedure. OpenAI’s retention guidance says deleting a chat does not delete files saved to Library, and users must view and delete saved files from Library. For shared folders, this separation is especially important because a file uploaded by a contributor into a shared folder belongs to the folder owner while the uploader remains recorded. If that contributor later loses folder access, OpenAI says the file remains in the owner’s folder and the contributor loses access; there is no implication that the file is deleted, transferred back, or made inaccessible to the folder owner.

This article explains how enterprise AI governance is evolving in 2026, comparing Microsoft Purview with OpenAI’s built-in compliance tools for risk management, data protection, and operational governance. The complete How Enterprise AI Governance Is Evolving in 2026: From Microsoft Purview to OpenAI’s Built-In Compliance Tools article provides the destination-specific detail for this section’s ChatGPT Retention and Compliance decision because it is appropriate for retention and compliance because the current article’s revocation and file-custody guidance needs a broader compliance-governance frame, while the other high-score retention result is about customer retention and is off-topic.

Inactive, expired, unsupported, and transient files need explicit handling

OpenAI distinguishes active Library files from files that are deleted, expired, or inactive for Enterprise Compliance API purposes. Once a Library file is deleted, expired, or inactive, OpenAI says it becomes unavailable immediately through the Library-specific Compliance API endpoints, although internal backups may retain it for up to 30 additional days subject to legal, security, and other exceptions. That creates a practical evidence rule: collect compliance evidence while the file is active if your organization has a legitimate need to preserve it.

Unsupported or transient files can expire separately, according to OpenAI’s retention guidance. Administrators should avoid building processes that assume every uploaded item becomes a durable Library record with the same retention behavior. A defensible process classifies each file as a durable Library file, a project or GPT attachment, a chat-associated file saved to Library, or a transient/unsupported item whose availability may expire independently.

For regulated teams, “inactive” should be treated as an operational state that can affect compliance availability, not as proof that the file has been permanently erased everywhere. OpenAI’s backup note means a file can be unavailable through compliance endpoints while still present in internal backups for a limited additional period, subject to exceptions. Security teams should therefore coordinate evidence preservation, account suspension, and deletion requests instead of sequencing them informally through chat messages or ad hoc owner requests.

Project and GPT files are not governed by the same checklist as Library folders

Files attached to custom GPTs and projects, including shared projects, have their own lifecycle. OpenAI states that those files are retained until the GPT or project is deleted, and then removed within 30 days subject to legal or security exceptions. A team that decommissions a department folder must still separately review custom GPTs and projects that may contain policy documents, product specs, customer notes, source excerpts, or other attachments.

The operational mistake is to search only shared Library folders during offboarding. A departing project owner may have maintained a shared project with files that are outside the folder inventory, or a custom GPT may contain knowledge files that continue to shape responses after the employee leaves. The correct control is an asset inventory that names Library folders, shared folders, shared projects, custom GPTs, and any known workflows that use uploaded files.

Deletion of a GPT or project should be governed by business approval, not by a hasty account cleanup step. If a shared project supports an active team workflow, deletion may interrupt work and start the file-removal clock described by OpenAI. If the files are obsolete or sensitive and no longer needed, retention policy and legal/security review should determine whether deletion is appropriate before the owner’s access is removed.

Joiner-mover-leaver checklist for Library sharing

A joiner-mover-leaver process should map people to business purpose before it maps them to folders. OpenAI’s Library sharing model allows sharing with named people as Viewer or Editor and across a workspace, so administrators and folder owners need a reasoned access model that can survive promotions, team moves, contractor endings, and incident response. The following checklist is a recommended governance workflow, not a statement that ChatGPT automatically enforces every step.

Joiner checklist: grant the minimum useful access

1. Identify the person’s work purpose. Assign access only to folders needed for the role, project, or onboarding path; do not use workspace-wide sharing as a substitute for role design.
2. Select Viewer or Editor deliberately. Use Viewer where the person only needs to reference files in conversations, and reserve Editor for contributors who are expected to add or change shared folder contents.
3. Explain custody before upload. Tell contributors that files uploaded into another person’s shared folder belong to the folder owner while the uploader is still recorded.
4. Classify files before first use. Require the joiner to follow the folder’s sensitivity, naming, and source rules before adding documents that may later be used by ChatGPT or Deep Research.
5. Record the business owner. Maintain a lightweight register naming the folder owner, steward, data category, approved audience, and review date.

Mover checklist: remove old access before adding new access

1. Inventory current shared access. Review folders shared directly with the person and any workspace-wide folders they use for their old function.
2. Separate access from files they uploaded. Removing access to a prior folder does not delete files they uploaded there if the folder belongs to someone else.
3. Ask the old folder owner to review contributed files. The owner should retain, delete, or reclassify files according to business need and retention policy.
4. Regrant access for the new role. Add Viewer or Editor access only after the new role’s purpose and sensitivity rules are clear.
5. Check projects and GPTs. Confirm whether the mover owns or maintains projects or custom GPTs whose files should remain, be reassigned operationally, or be decommissioned under policy.

Leaver checklist: preserve what the business needs and remove what the person no longer needs

1. Freeze the access inventory. List shared folders, workspace-shared items, projects, custom GPTs, and known Library files relevant to the departing person’s work.
2. Remove access to folders the person should no longer use. For shared folders owned by others, this blocks future access but does not delete files already uploaded into those folders.
3. Review folders owned by the leaver. Identify files that must remain available to the team, files that should be deleted under policy, and files that require legal or security preservation.
4. Do not assume automatic ownership transfer. OpenAI’s cited Library sharing notes document folder-owner custody but do not document an automatic ownership-transfer feature.
5. Collect compliance evidence before destructive steps. Where Enterprise Compliance API workflows apply, remember that deleted, expired, or inactive Library files become unavailable immediately through Library-specific endpoints.
6. Review chat cleanup separately. Deleting or archiving chats does not dispose of saved Library files, and archiving does not delete chats.
7. Review GPT and project files separately. Files attached to projects or custom GPTs persist until the project or GPT is deleted, then follow the deletion timing and exceptions described by OpenAI.

This guide explains OneNote governance for ChatGPT and Codex, including notebook scope, decision capture, safe updates, and copy verification for the beta OneNote plugin. The complete OneNote in ChatGPT and Codex Governance Guide: Notebook Scope, Decision Capture, Safe Updates, and Copy Verification article provides the destination-specific detail for this section’s OneNote Governance Guide decision because it is the exact subject match for the OneNote governance marker and directly complements library-sharing governance with notebook-scope and knowledge-work controls.

Ownership-transfer checklist without assuming automatic transfer

Because OpenAI’s Library sharing note defines folder-owner custody but does not document an automatic transfer mechanism, ownership changes should be handled as a controlled governance event. The goal is to make the intended business owner responsible for ongoing access decisions and retention, while avoiding undocumented assumptions about metadata preservation, uploader attribution, file movement, or automatic reassignment.

1. Name the successor owner. Choose a person or function accountable for access review, deletion decisions, sensitivity classification, and business continuity.
2. Inventory the source folder. Record file names, business purpose, sensitivity, known uploader attribution, current shared audience, and whether each file is still needed.
3. Identify legal, security, or retention constraints. Do not delete, recreate, or reorganize files that are subject to a legal hold, security investigation, records requirement, or workspace retention rule.
4. Select a permitted transition method. If the product and policy allow a direct administrative or user-driven method, use it and verify the resulting owner. If not, create a controlled replacement under the intended owner using available, authorized operations, and document that this was a replacement or migration rather than an automatic ownership transfer.
5. Preserve attribution facts where needed. Since OpenAI records the uploader for files added to shared folders, governance records should not overwrite the distinction between the former uploader and the new operational owner.
6. Validate access after transition. Confirm that the successor owner can manage the folder, approved collaborators retain the intended Viewer or Editor access, and the departing or moving person no longer has access unless explicitly approved.
7. Retire the old folder only after verification. Do not delete the prior folder or project until the successor confirms that required files are available, obsolete files are handled under policy, and compliance evidence has been preserved where required.

Deep Research raises the cost of poor retention hygiene

OpenAI’s Deep Research guidance says research can use uploaded files and, where available and authorized, supported connected apps; starting Deep Research does not grant additional access to files or apps. That boundary protects against permission bypass, but it also means stale or over-shared Library files can still become part of a research task if the user already has access and includes or relies on them. Retention hygiene therefore affects research quality as well as confidentiality.

Teams should define a pre-research file check for sensitive folders. Before asking Deep Research to synthesize a folder, the requester should verify that the files are current, approved for the audience, and not retained merely because a former contributor uploaded them months earlier. Citations and source links improve traceability, but OpenAI warns that they do not guarantee correctness, so users must inspect the underlying files and sources before decisions, external publication, or high-stakes use.

Operational rule: access revocation protects the future, deletion handles disposal, retention policy governs required preservation, and ownership determines who is accountable. Treat those as four separate controls in every offboarding and folder-transfer workflow.

Retention exceptions should be documented before cleanup begins

OpenAI’s retention descriptions repeatedly include exceptions for legal, security, and related circumstances. Administrators should translate that into a written decision gate: before deleting chats, Library files, projects, GPTs, or folders associated with a leaver, determine whether the content is subject to litigation hold, internal investigation, security review, contractual retention, regulated records policy, or workspace retention settings. The gate matters because deletion actions may remove user-visible access or Compliance API availability before all stakeholders have preserved what they are authorized to preserve.

A practical evidence log should capture the folder owner, uploader where visible or recorded, access list, retention decision, deletion decision, approver, and date. The log should not claim that a file has been permanently purged from every backup or exception path unless the organization has a separate verified basis for that statement. OpenAI’s guidance permits standard deletion timelines to be modified by de-identification, disassociation, backup retention, and legal or security exceptions, so governance records should state the action taken rather than promise an outcome the workspace cannot independently prove.

Operating charter for Library sharing governance

A workable Library governance charter should state that sharing is an access decision, folder custody is an ownership decision, uploader attribution is an audit fact, and deletion is a separate retention action. OpenAI’s release notes establish that Library users can share files and folders with named people as Viewer or Editor, share across a workspace, review or change access, remove access, and use shared items from “Shared with me” in conversations. The same notes also establish a custody rule that matters operationally: files uploaded into a shared folder belong to the folder owner, while the uploader remains recorded.

• Purpose clause: Shared folders exist for defined business work such as team research, client deliverables, policy drafting, or reusable knowledge, not for unmanaged personal storage.
• Least-privilege clause: Viewer is the default role for reference use; Editor is reserved for people who must add, change, or curate materials in the folder.
• Custody clause: The folder owner is accountable for files that remain in the folder, including files uploaded by other contributors.
• Attribution clause: The uploader identity should be preserved in review records so teams can resolve provenance, rights, and cleanup questions.
• Revocation clause: Removing someone’s access is not a deletion event and must not be treated as cleanup of files they uploaded.
• Retention clause: Chat deletion, chat archiving, and Library deletion are different actions; deleting a chat does not delete Library files saved from or used in that chat.
• Research-use clause: Deep Research workflows should use only sources that the user and workspace are authorized to access, and citations must be verified before external or high-stakes use.

Operational rule: Do not approve any shared Library folder unless it has a named owner, a business purpose, a sensitivity classification, a minimum access role, a review cadence, and a deletion or retention procedure that is separate from access revocation.

RACI for shared Library folders

This article explains how eligible ChatGPT Business and Enterprise users can share a ChatGPT Site with named external viewers without making it public or adding viewers to the workspace. The complete How to Share a ChatGPT Site with External Clients Without Making It Public article provides the destination-specific detail for this section’s External Sharing Security Guide decision because it is the best match for external sharing security because it focuses on named external access, avoiding public exposure, and client-review sharing patterns that parallel safe library-sharing practices.

Sharing decision tree for owners and administrators

1. Identify the audience. If the audience is a named person or an internal workspace population, continue. If the proposed pattern depends on anonymous public-link sharing, stop; the September 9 release notes describe named-person sharing and workspace sharing, not public links.
2. Identify the action needed. If the person only needs to reference files in conversations, assign Viewer. If the person must add or maintain files in the folder, consider Editor after reviewing sensitivity and custody implications.
3. Check folder custody impact. If an Editor may upload files, confirm that the folder owner accepts custody because OpenAI states that files uploaded into a shared folder belong to the folder owner, while the uploader remains recorded.
4. Check sensitivity and rights. If files include customer data, confidential research, licensed content, privileged material, or regulated data, require a data-owner review before sharing and before using the files in Deep Research outputs.
5. Check retention expectations. If collaborators expect access removal to delete their uploaded files, correct that assumption before granting Editor access. OpenAI’s documented behavior is that if a contributor loses folder access, the file remains in the owner’s folder and the contributor loses access.
6. Check cleanup path. If there is no owner able to review, delete, or retain files under the applicable policy, do not share the folder until custody and records responsibilities are assigned.
7. Approve and document. Record the folder name, owner, purpose, access list, role rationale, sensitivity classification, and next review date in the team’s governance record.

Incident response for oversharing

An oversharing incident occurs when a file or folder is made available to people who do not have a business need, when Editor access is granted where Viewer would have been sufficient, when workspace-wide sharing exposes sensitive material too broadly, or when a shared file is used in a downstream research or decision artifact without proper authorization. The first response is containment, not deletion, because deletion may conflict with investigation, retention, or legal obligations.

1. Contain access. Remove unnecessary named users or workspace-wide access, reduce Editor roles to Viewer where editing is no longer required, and preserve the folder owner’s ability to inspect the contents.
2. Freeze uncontrolled changes. Ask Editors to stop uploading, modifying, or deleting files until the owner and security or compliance reviewer classify the incident.
3. Record the observed state. Capture the folder owner, current access list, roles, file names or identifiers available to your organization, uploader attribution where visible, and the time the issue was found.
4. Classify exposure. Determine whether the material was internal-only, confidential, regulated, customer-provided, third-party licensed, privileged, or intended for public release.
5. Review downstream use. Identify whether shared files were used in conversations, Deep Research tasks, reports, presentations, spreadsheets, or other artifacts, and require source verification before those artifacts continue circulating.
6. Decide retention before deletion. Consult the applicable workspace retention policy and legal or security requirements before deleting files, because OpenAI’s retention documentation includes policy-based handling and legal or security exceptions.
7. Remediate and notify. Remove excess access, delete or retain files according to policy, notify affected data owners, and document the control change that prevents recurrence.

Export and deletion verification

Verification should prove two different things: what evidence the organization preserved for governance review, and what content was actually removed from active Library access. Do not describe a governance export, a screenshot, an access report, or a compliance-system record as deletion. An exported record is evidence; deletion is the removal of the file or chat object through the applicable ChatGPT or workspace procedure.

Quarterly audit evidence package

A quarterly audit should produce evidence that the organization separated permission review from content cleanup. The access review answers “who can see or edit this folder”; the custody review answers “who owns the folder and files”; the retention review answers “what should be deleted, retained, or escalated”; and the research review answers “whether shared files were used in outputs that require citation, rights, or sensitivity checks.”

• Folder inventory: list of shared folders, owners, purpose statements, sensitivity classifications, and review dates.
• Access inventory: named users, workspace-wide shares where used, Viewer and Editor assignments, and role justification for Editors.
• Custody inventory: files uploaded by contributors into owner-controlled folders, with uploader attribution where available to reviewers.
• Revocation log: users removed during the quarter, reason for removal, date of removal, and confirmation that files were separately reviewed.
• Deletion log: Library files deleted, chat threads deleted where relevant, and exceptions where deletion was paused for workspace retention, legal, or security reasons.
• Deep Research review: research artifacts that relied on shared files, citation checks performed, source-access assumptions reviewed, and human approvals for external or high-stakes use.
• Exception register: folders without owners, broad workspace shares, long-running Editor assignments, unresolved cleanup questions, and remediation owners.

Rollout scorecard for Library sharing

What the release notes do not establish

The September 9 release notes establish a specific Library sharing model, but they do not establish public-link sharing. Teams should not design policies, training, or external collaboration workflows around anonymous links unless OpenAI separately documents and enables that capability for the relevant environment. For this governance guide, the safe boundary is named people, workspace sharing, Viewer and Editor roles, access review, access removal, and “Shared with me” use in conversations.

The release notes also do not establish automatic deletion when access is revoked. OpenAI states that if a contributor uploads a file to another person’s shared folder and later loses access, the file remains in the owner’s folder and the contributor loses access. That means revocation is a permission control, not a records cleanup control, not an ownership transfer, and not proof that copies, outputs, or retained files have been removed under the applicable retention policy.

Conclusion: govern Library sharing as custody, not convenience

Library sharing becomes safe enough for serious work only when teams treat it as a custody system with collaboration features, not as a casual file handoff. The practical rule is simple: grant the minimum role, name the accountable owner, warn Editors about custody before upload, document revocation separately from deletion, and verify any research artifact that uses shared files. This operating model keeps OpenAI’s documented boundaries intact while giving administrators, security teams, and knowledge workers a repeatable way to collaborate without confusing access, ownership, attribution, retention, and deletion.

Useful Links

• ChatGPT release notes
• Chat and file retention policies in ChatGPT
• Deep Research in ChatGPT