Identifying Appropriate vs Inappropriate AI Use Cases

Audience: professionals who use Claude at work, no coding or API knowledge needed. Verified = checked against Anthropic's documentation or help center on 2026-10-16. General background = standard regulatory or industry knowledge, not an Anthropic statement. Design guidance = widely taught practice.

This note is design guidance anchored in the review-level thinking from D2.1. Always apply your organisation's policy first.

Good fits

• Drafting, editing and summarising where you review the result.
• Brainstorming, outlining and planning.
• Explaining or simplifying material you can check.
• Extracting and organising information from documents you provide, with verification.
• First-pass analysis with your own data (allowed tools), checked by someone who understands it.
• Coding help, formulas, templates that you test.
• Learning a topic, then verifying key facts.

Use with care (needs strong review and controls)

• Client-facing or published content.
• Anything with numbers, citations, legal or regulatory references.
• Material with personal or confidential data (only in approved arrangements, D6.1).
• Advice in specialist fields (legal, medical, financial) that a qualified professional must own.
• Summaries that people will rely on instead of reading the source.

Do not use (or only as policy explicitly allows)

• Final decisions about people (hiring, firing, pay, credit, discipline, benefits) without meaningful human judgement.
• Safety-critical decisions (medical dosing, engineering sign-off, emergency response) without qualified human control. (Anthropic's own prompting guide includes cautionary examples for high-stakes areas such as medication dosing.)
• Tasks that require guaranteed accuracy you cannot verify.
• Regulated or classified data in non-approved tools.
• Deceiving, impersonating, harassing or manipulating people.
• Illegal activity or anything that breaks licences or contracts.
• Passing off AI output as your own verified expertise.

A fast screening test

1. Harm if wrong? High means more review or do not use.
2. Can I verify the output? If not, think twice.
3. Is the data allowed here?
4. Does a person remain accountable?
5. Does policy or law restrict it?
6. Does AI actually help (time, quality), or is it adding risk for little gain?

Exam traps

• "It's just a draft" for content that will be sent unchecked.
• Using AI for a task whose output you cannot evaluate.
• Assuming "AI said so" is a defence.
• Confusing "allowed" with "wise".