A clinic wants Claude to draft letters advising patients on treatment options. A nurse will read and sign each letter before it is sent. Under Anthropic's Usage Policy, how should this be assessed?

The principle — the policy's high-risk provisions turn on the effect a use has on people, so the analysis runs through who is affected and who checks the output, never through which product produced it.

Why the key is correct — Anthropic's Usage Policy states that when using its products or services to provide advice, recommendations, or in subjective decision-making directly affecting individuals or consumers, a qualified professional in that field must review the content before it is disseminated or finalised. Letters advising patients on treatment options are advice directly affecting individuals, so the obligation plainly attaches. It is also plainly satisfied: a nurse is a qualified professional in the relevant field, and the review happens before the letter is sent rather than after. The right conclusion is therefore neither that the clinic must abandon the idea nor that nothing is required — the obligation attaches and the described process discharges it.

Why the others are wrong — treating the category as prohibited misreads a conditional obligation as a ban, and would stop a clinic from doing something the policy expressly contemplates. Treating any human sign-off as sufficient drops the qualification requirement, which is the part doing the work; an administrator forwarding the same letter would not meet it. And confining the policy to developer use mistakes where the drafting happens for who is affected by the result, which is what the provision is written around.

Remember this — ask who is affected, then ask whether a qualified person in that field checks it before it leaves.

Sources — Anthropic's Usage Policy.