Three responsibilities apply every time: use only approved tools and information, review the result, and remain accountable for the final decision or communication.

AI can accelerate work, but it can also produce incorrect information, expose data through careless use, or reproduce bias. Use it as an assistant—not as an unquestioned authority or the final decision-maker.

Before you enter information

• Confirm that the AI tool and account are approved for SBS work.
• Follow the SBS Policy for AI and institutional information.
• Provide only the minimum information needed for the task.
• Remove names, identifiers, credentials, access links, and confidential details when they are not necessary.
• Check sharing settings before connecting a drive, mailbox, calendar, browser, or other service.
• When uncertain, pause and contact Technology Services or the appropriate data owner.

Use the minimum necessary access

When ChatGPT Work or Codex can act on files, applications, or the web, choose the narrowest permission mode that completes the task. For most local work, OpenAI recommends starting with Ask for approval. Review permission requests and confirm the target before approving an action.

• Examples include brainstorming with non-sensitive context, revising public text, creating a blank template, or explaining a general concept.
• Examples include using internal files, generating analysis that informs a decision, drafting communications about individuals, or allowing an agent to edit files or use connected services.
• Stop when the task involves credentials, restricted or regulated information, an unclear data classification, a high-impact decision about a person, or an external action you are not authorized to take.

Review the output before use

• Facts: verify names, dates, numbers, quotations, calculations, and policy statements.
• Sources: open cited sources and confirm that they support the exact claims.
• Privacy: check the prompt, output, file metadata, comments, and sharing permissions.
• Fairness: look for unsupported assumptions, stereotypes, exclusions, and unequal effects.
• Audience: confirm that tone, accessibility, reading level, and context are appropriate.
• Accountability: make the final judgment yourself and follow required review or approval processes.

Special caution for people-related decisions

Do not let an AI system make an unreviewed decision about admission, grading, accommodations, employment, discipline, performance, access, or eligibility. These contexts can involve policy, legal obligations, incomplete data, and significant consequences. Use the established SBS process and involve the responsible office.

If something goes wrong

1. Stop the task and do not share or act on the output.
2. If information may have been exposed, preserve the relevant details without reposting sensitive content.
3. Contact Technology Services through the normal support channel.
4. Follow any incident-reporting, records, privacy, or supervisory requirements that apply.

Related guides

• Research current information with AI and verify the sources
• Get started with Codex safely
• Review, correct, and improve an AI response

Official guidance

See OpenAI’s Permissions and Web search guides. SBS policy and instructions take precedence for institutional work.