Authorized inputs
Users enter information only when adequate authority, consent, institutional permission, contractual permission, and system approval support that use. Each project identifies data ownership, stewardship, minimum necessary access, approved tools, and release authority.
Restricted inputs
Protected health information, identifiable patient or research-participant information, confidential datasets, authentication secrets, restricted manuscript text, export-controlled information, proprietary third-party material, and information supplied without adequate authority remain within separately approved secure systems and processes.
AI-supported work
AI may support ideation, organization, editing, coding, analysis planning, visualization, translation, and other approved tasks. Material use receives documentation appropriate to the project, source, risk, venue, and applicable disclosure requirements.
Human accountability
Qualified human review governs scientific interpretation, factual accuracy, citation verification, clinical or professional judgment, methods, code, authorship, ethical decisions, and publication claims. Each accountable author or decision owner remains responsible for the final work.
Accuracy and evolution
AI output may contain inaccuracies, omissions, bias, outdated information, fabricated citations, unsuitable context, or unstable results. Systems, models, capabilities, and limitations evolve. Verification records should identify the source, review, correction, and final decision.
Data lifecycle
Project records identify collection, source, purpose, recipients, legal or institutional basis, retention, security, transformation, deletion, de-identification, export, publication, and legal-hold requirements.
Aggregation and de-identification
Aggregation and de-identification require documented transformations, access restrictions, reasonable assessment of re-identification risk, purpose limits, quality review, and approval for release.
Research and publication
AI-assisted research follows the approved protocol, data-use conditions, institutional policies, journal rules, authorship standards, and disclosure expectations. AI systems receive no authorship designation or independent decision authority.