Collaboration: answer first
An approved researcher can invite up to four collaborators from the same institution. Each collaborator must verify affiliation and counts toward the program total. Treat the shared workspace as an access surface, not permission to every dataset, system, participant record, or result.
Define the team before inviting
| Role | Owns |
|---|---|
| Principal researcher | Question, scope, approvals, final scientific claim |
| Data steward | Source rights, minimization, access, retention, deletion |
| Analysis owner | Code, environment, methods, tests, reproducibility |
| Reviewer | Independent source, statistical, safety, and citation review |
| Workspace admin | Identity, membership, settings, incidents, offboarding |
One person can hold multiple roles, but every responsibility needs a named owner.
Keep a reproducible project record
Store approved source versions, data lineage, code, environment, prompts, model and tool context, outputs, failures, reviewer decisions, and the final accepted artifact in the lab’s controlled system. Workspace chat history alone is not a complete research record.
Use least-privilege connectors and repositories. Re-authorize when the data class, source system, action, collaborator, destination, or research purpose changes.
Frequently asked questions
Is a team limited to five people?
The launch describes one approved researcher plus up to four invited collaborators. Verify current limits for the actual program workspace.
Does ChatGPT Edu change collaboration rules?
OpenAI says access is coordinated through the institution’s ChatGPT Edu workspace. Local admin, identity, retention, and tool policies still apply.
Can the team publish an AI-generated conclusion directly?
No. Qualified researchers must verify evidence, code, statistics, citations, interpretation, safety, and authorship or disclosure requirements.
Official sources
Source check: August 5, 2026.