Harvard University

HGSE states that, unless the instructor specifies otherwise, using generative AI to create all or part of a course assignment and submit it as one’s own violates its Academic Integrity Policy. Permissible learning uses include clarification, brainstorming and contextual scenarios; students must discuss uncertain uses with the instructor before use.

FAS Office of Undergraduate Education requires faculty to inform students of the policies governing generative AI use in class. It recommends posting the policy on Canvas. This communication requirement does not choose an AI permission policy for the course.

For permitted GenAI use in HGSE assignments, students must acknowledge and document the tools used, prompts if applicable, and how outputs were integrated. Direct citations must use proper citation format.

Harvard meeting guidance says AI assistants should not be used except approved tools with contractual protections or applicable approved pilots/accommodations. Participants must be informed and all must consent before recording/transcription begins. The guidance says not to use assistants when participants object (with a disability-accommodation consultation route), and to disable them in meetings involving Level 3 or higher information. Summaries must be reviewed before limited sharing; retention and litigation holds also apply.

HUIT’s general-use comparison applies Level 3-or-below approval and no-vendor-training protection only when logged in with a Harvard account. It does not approve Level 4 data. HMS separately lists special-request Level 4 tools and a Level 2 ceiling for automated meeting assistants. Product access, data approval, DUA/sponsor limits and assessment permission must be checked separately.

Harvard guidance says Level 2 and above confidential data should not be entered into publicly available generative AI tools and must only be entered into tools assessed and approved for that use. HUIT-listed managed general-use tools are approved up to Level 3 with a Harvard account and are not approved for Level 4. Separate HMS Level 4 research/development services require their own access and data-use conditions.

FAS OUE guidance requires documented student permission before faculty put original student content into a generative AI tool and says students should be informed of the risks. Its page also contains a broad no-confidentiality statement. HUIT separately describes contractual privacy protections for approved managed services; that data approval does not waive FAS student-content consent or course rules.

Gemini Notebook (formerly NotebookLM; Basic) is included in Harvard's Basic Google Workspace tier for eligible account holders. Gemini Notebook is the current name of NotebookLM; Pro includes expanded Notebook limits. HUIT lists specific eligible Schools and user groups, with local approval for some Google Workspace account requests. Standard Gemini products are approved up to Level 3 confidential data, while agent sessions require active monitoring and connectors require separate review. Access is separate from assessment permission.

Gemini Notebook (formerly NotebookLM; Pro) is included in Harvard's paid Google AI Pro for Education tier, available by request. Gemini Notebook is the current name of NotebookLM; Pro includes expanded Notebook limits. HUIT lists specific eligible Schools and user groups, with local approval for some Google Workspace account requests. Standard Gemini products are approved up to Level 3 confidential data, while agent sessions require active monitoring and connectors require separate review. Access is separate from assessment permission.

HMS lists AI Sandbox, Claude, ChatGPT Edu, Microsoft Copilot Premium, Firefly, HMS Azure AI, HUIT API Portal and Longwood Cluster with different eligibility, training, data levels and costs. HMS provides one of Claude, ChatGPT Edu or Copilot Premium per eligible person. HMS Azure AI is a Level 4 service for HMS IT/project developers, and Longwood is a separate Level 4 research environment; this is not blanket Level 4 access for ordinary students.

HMS responsible-use guidance advises users to verify AI outputs, follow applicable faculty/student handbook rules, be transparent about acceptable use and cite AI contributions appropriately. It separately requires attention to confidential data, regulations, sponsor requirements and DUA restrictions; it does not establish assessment permission.

HMS Academic and Research Integrity guidance says AI tools cannot be listed as authors on papers. Authors should be transparent about using AI tools and explain how the tools were used.

Harvard guidance recommends that faculty clearly communicate permitted AI uses, if any, in classes and academic work; students are encouraged to ask their instructors for clarification. The applicable School and course rules determine permission.

Harvard guidance requires vendor generative AI tools to be assessed for risk by Harvard Information Security and Data Privacy before use in Harvard work. Users considering tools not currently offered should contact HUIT before procurement. Enabled connectors and new features may need their own review.

HGSE forbids making personal recordings of course meetings, with or without AI integrations; students needing recordings should discuss them with the instructor. Uploading substantial course content is only allowed through the Harvard-approved AI Sandbox, accessed through individual courses. Assistive-technology AI use must be coordinated with the Office of Student Affairs.

AWS Bedrock via Harvard API Platform is available through Harvard's API Platform for eligible users. The platform offers self-service AI API access, including pay-as-you-go or limited-access credit-redemption options; some individual APIs have limited eligibility stated in their documentation. API/developer services are approved up to Level 3 confidential data. Account access is separate from course, research-use and feature-specific approval.

Harvard-managed ChatGPT Edu is available by request at a monthly cost, with School or Unit approval and the service's listed eligibility. Codex is included with Harvard ChatGPT Edu licences. ChatGPT Chat and Codex are approved up to Level 3 confidential data; ChatGPT Work requires active monitoring of autonomous sessions for that data level. API keys require separate API Platform access, and connectors require security/privacy approval. Institutional access does not grant assessment permission.

Harvard Claude is available in paid Standard and Premium tiers, with requests reviewed and approved by Schools and Units. Licences include Chat, Code, Cowork and Science; API keys use a separate API Platform path. Chat and Code are approved up to Level 3 confidential data; Cowork requires active monitoring of autonomous sessions. Connectors require security/privacy approval. Tool access does not grant assessment permission.

Harvard-managed ChatGPT Edu is available by request at a monthly cost, with School or Unit approval and the service's listed eligibility. Codex is included with Harvard ChatGPT Edu licences. ChatGPT Chat and Codex are approved up to Level 3 confidential data; ChatGPT Work requires active monitoring of autonomous sessions for that data level. API keys require separate API Platform access, and connectors require security/privacy approval. Institutional access does not grant assessment permission.

Google Gemini (Basic) is included in Harvard's Basic Google Workspace tier for eligible account holders. Gemini Notebook is the current name of NotebookLM; Pro includes expanded Notebook limits. HUIT lists specific eligible Schools and user groups, with local approval for some Google Workspace account requests. Standard Gemini products are approved up to Level 3 confidential data, while agent sessions require active monitoring and connectors require separate review. Access is separate from assessment permission.

Google Gemini (Pro) is included in Harvard's paid Google AI Pro for Education tier, available by request. Gemini Notebook is the current name of NotebookLM; Pro includes expanded Notebook limits. HUIT lists specific eligible Schools and user groups, with local approval for some Google Workspace account requests. Standard Gemini products are approved up to Level 3 confidential data, while agent sessions require active monitoring and connectors require separate review. Access is separate from assessment permission.

Google Gemini API via Harvard API Platform is available through Harvard's API Platform for eligible users. The platform offers self-service AI API access, including pay-as-you-go or limited-access credit-redemption options; some individual APIs have limited eligibility stated in their documentation. API/developer services are approved up to Level 3 confidential data. Account access is separate from course, research-use and feature-specific approval.

HUIT's feature comparison lists Workspace Studio with Google Gemini Basic. This is a Gemini feature entry, not a separate blanket tool licence. Basic Google Workspace access is free for eligible users; Google AI Pro for Education requires a request at additional cost. School/account eligibility and enabled-feature/connector approvals still apply. Tool access does not grant assessment permission.

HUIT's feature comparison lists Workspace Studio with Google Gemini Pro. This is a Gemini feature entry, not a separate blanket tool licence. Basic Google Workspace access is free for eligible users; Google AI Pro for Education requires a request at additional cost. School/account eligibility and enabled-feature/connector approvals still apply. Tool access does not grant assessment permission.

Harvard AI Sandbox is a free institutional multi-model service for the School/user groups listed on its service page. It is approved up to Level 3 confidential data and states that input data will not train vendor models. HGSE access to the Sandbox for course-content uploads is through individual courses; HMS access/training conditions are School-specific. Access does not grant assessment permission.

AI APIs and Developer Tools via Harvard API Platform is available through Harvard's API Platform for eligible users. The platform offers self-service AI API access, including pay-as-you-go or limited-access credit-redemption options; some individual APIs have limited eligibility stated in their documentation. API/developer services are approved up to Level 3 confidential data. Account access is separate from course, research-use and feature-specific approval.

Harvard Microsoft Copilot Chat is included at no extra cost for eligible Harvard Microsoft 365 users, who sign in with HarvardKey. The Premium tier is available by request at additional cost and includes Microsoft 365 application integration. The service lists School/user eligibility. Institutional access and Level 3 approval for named managed products do not grant assessment permission; meeting use and connector approvals have additional conditions.

Harvard Microsoft Copilot Chat is included at no extra cost for eligible Harvard Microsoft 365 users, who sign in with HarvardKey. The Premium tier is available by request at additional cost and includes Microsoft 365 application integration. The service lists School/user eligibility. Institutional access and Level 3 approval for named managed products do not grant assessment permission; meeting use and connector approvals have additional conditions.

OpenAI Direct via Harvard API Platform is available through Harvard's API Platform for eligible users. The platform offers self-service AI API access, including pay-as-you-go or limited-access credit-redemption options; some individual APIs have limited eligibility stated in their documentation. API/developer services are approved up to Level 3 confidential data. Account access is separate from course, research-use and feature-specific approval.

Harvard guidance says users are responsible for any content they publish or share that includes AI-generated material, which may be inaccurate, fabricated or copyrighted.