What you’ll be able to do
- Sort tasks by stakes and reversibility into appropriate, review-required and inappropriate
- Rewrite an inappropriate request into an appropriate supporting role for Claude
- Explain who holds accountability for a decision Claude helped with
What you need to know
The question is not “can Claude do it?” but “who must own it?”
Claude can produce a plausible answer to almost anything: a ranking of job candidates, a reading of a contract clause, a view on a symptom, an investment recommendation. Capability isn't the test. The test is what happens if the output is wrong, who is affected, and who has to be accountable for the outcome. A useful way to sort tasks:
- Appropriate: low-stakes, easy to check, reversible — brainstorming campaign names, tidying meeting notes, drafting an internal FAQ, outlining a lesson plan.
- Appropriate with human review: the output matters, but a qualified person checks it before it's used — a customer-reply draft, a policy summary, a first-pass analysis of survey comments, a draft contract clause for the lawyer to review.
- Inappropriate as Claude's call: the decision materially affects someone's rights, livelihood, health or finances, or requires professional accountability — deciding who gets hired or let go, approving a loan, giving a final diagnosis, signing off a legal position.
High-stakes decisions about people
Decisions about individuals — hiring, promotion, performance ratings, discipline, termination, admissions, benefits eligibility — carry legal and ethical obligations, and they can encode bias in ways that are hard to see. Claude may help with the process: drafting a structured interview guide, summarizing what a candidate wrote in their own application, organizing a manager's notes into the company's review template. It must not be the one deciding, ranking or scoring people as the basis for the outcome. A person with authority makes the decision, using criteria they can explain and defend.
Final legal, medical and financial judgments
Claude can explain concepts, summarize documents, prepare questions to ask a professional, and draft material for a professional to review. It cannot be the source of a final legal opinion, a diagnosis or treatment decision, or a binding financial judgment. Those require a qualified, accountable professional who knows the full facts and answers for the advice. If a scenario has someone acting on Claude's answer in place of that professional, it's the wrong answer.
Accountability can’t be delegated to a tool
Anything requiring accountability a human must hold — a signature, a sign-off, a regulatory attestation, a decision someone may need to explain or appeal — stays with a person. Adding "AI-generated, please verify" to the bottom of an output doesn't shift responsibility; the person who uses it is still accountable. Nor is review "at some point later" enough: the check has to happen before the output affects anyone.
Warning signs in a request
Some phrasings should make you pause and reconsider the task before you start:
- "Decide," "pick," "rank," "approve" or "reject" applied to people, applications or claims.
- "Just tell me if…" for a legal, medical, tax or safety question that will be acted on directly.
- No reviewer named for output that will reach customers, patients, students or the public.
- Speed as the justification for skipping a step that exists to protect someone ("we need this by tonight, so let's let Claude choose").
None of these automatically rules Claude out. They signal that you need to separate the supporting work Claude can do from the decision a person must make, and to check your organization's policy on that type of use.
Redesign rather than refuse
The opposite mistake is avoiding Claude entirely whenever a topic sounds sensitive. Most risky requests have a safe, valuable version. "Rank these 40 applicants" becomes "summarize each application against the published criteria, in a consistent format, so the panel can assess them" — with the panel making every decision, and with personal data handled under your organization's policy (lesson 6.2). The exam usually rewards this middle path: keep Claude in a supporting role and the human in the decision.
Key concept
Claude can support nearly any task, but when a decision seriously affects a person or requires professional accountability, a qualified human makes and owns it.
Exam options typically include "let Claude decide" (wrong), "don't use Claude at all" (usually wrong), and "use Claude to prepare or organize, with a qualified person deciding" (usually right). Judge by the stakes, not by how polished the output is.
Practice scenario
Work it through, then open this
This is a high-stakes decision about people, with legal exposure and real potential for bias — and Claude would be the one making it. A well-reasoned shortlist doesn’t change that. The appropriate response is to decline that use and redirect: the decision stays with managers and HR, using documented criteria and the organization’s process (often with legal or HR-policy review). Claude could help in a supporting role, such as drafting the communication plan or a checklist of process steps, provided data handling follows policy. Refusing all involvement or simply adding a “please verify” note would both miss the point.
Hands-on exercise — Sort and redesign real tasks
Beginner · 15 min
What you’ll learn
- Classifying tasks by stakes and accountability
- Finding the supporting role in a high-stakes task
- Stating who decides
-
List ten tasks from your role (or your team’s). Ask Claude to help sort them into appropriate, appropriate with review, and inappropriate as Claude’s decision — then review its sort yourself.
- Why: The sort is a judgment call, and practising it on your own work builds the instinct.
- You should see: Mostly agreement, plus a few cases where you, knowing the stakes, would move a task to a stricter category.
-
Take one task in the “inappropriate” group and ask Claude to suggest a supporting role it could play while a person keeps the decision.
- Why: Most high-stakes work has a safe, useful assist.
- You should see: A reframed task such as “prepare,” “organize” or “draft for review,” rather than “decide” or “rank.”
-
For that reframed task, write one sentence naming who reviews the output, when, and who is accountable for the final decision.
- Why: Naming the accountable person makes the boundary concrete.
- You should see: A named role and a checkpoint that happens before the output affects anyone.
Exam traps
Letting Claude make the final call on hiring, firing, promotion or discipline
Decisions about people carry legal, ethical and bias risks that require an accountable human decision-maker.
Treating Claude’s answer as final legal, medical or financial advice
Claude can explain and prepare; a qualified professional gives the final judgment and answers for it.
Assuming a task is safe because the output will be reviewed ‘at some point’
Review has to happen before the output affects anyone, by someone qualified to catch errors.
Refusing to use Claude at all for any task that touches a sensitive area
Most sensitive tasks have a valuable supporting role. Redesign the task rather than abandon it.
Believing a disclaimer in the output transfers accountability
The person who uses the output remains responsible for it, label or no label.
Judging appropriateness by how good the output looks rather than by the stakes
A polished answer can still be wrong or biased. Stakes and accountability decide, not quality of the prose.
Sources
- Claude Help Center
- Claude Docs
- Anthropic, Claude Certified Associate – Foundations Exam Guide v1.0 (July 2026)