Content Quality: Well structured News piece (861 words, range 400-1200). Overview, problem, mechanism, platform differences, webhooks, What We Don't Know, and a cross-link. Dense but accurate; every paragraph maps to one of the three cited pages.
Source Verification: Snapshots read (gunzipped, text-extracted): source-0.html.gz (release notes, 200): the October 20, 2026 entry contains the exact quoted phrase 'tell Anthropic which of your end customers each API request is for' (the page text reads 'so you can tell Anthropic which of your end customers each API request is for'), 'a safety enforcement action can then apply to one end customer instead of your whole organization', SDK and ant CLI support for /v1/user_profiles, organization webhooks in beta with the webhooks-2026-10-07 header, two event types, 'To request access, contact your Anthropic account team', and 'Inference hooks are out of beta for Claude Enterprise organizations'. source-1.html.gz (user profiles docs, 200): confirmed POST /v1/user_profiles, uprof_ ID, anthropic-user-profile-id header, token counting/Message Batches/OpenAI SDK compat with batch applying to every request, access_type application (default) vs passthrough (not default, one per company), every field optional, no conversation content, 'Anthropic doesn't verify the details', name_hash/email_hash and no field for the address itself, 'Profiles aren't billable and don't affect pricing', separate organization advice, 400 invalid_request_error on malformed/other-org ID, Claude API / Claude Platform on AWS / Foundry hosted on Anthropic can create profiles, Bedrock and Google Cloud lack the endpoints, Bedrock user_profile_id body field, any well-formed ID accepted there, 'goes through unattributed', metadata.user_id advice, delete semantics (trust and safety records kept, up to about 10 minutes). The quote 'Anthropic runs automated safety classifiers over API traffic' is verbatim. source-2.html.gz (organization webhooks docs, 200): beta status and header webhooks-2026-10-07 confirmed; the note reads 'Event delivery isn't live: Anthropic doesn't send the two events, so an endpoint you register receives nothing.' which matches the article's quotation exactly; event names, data.id, enac_ prefix, retrieve-profile for restriction/ban type, Standard Webhooks, 2xx within 5 seconds, retries 'a few times over the next few minutes', 32 endpoints, access via account team all confirmed. Live re-fetch of all three URLs on 2026-10-21 (WebFetch) matched the snapshots: no changes, no newer release-notes entry, the 'Event delivery isn't live' note still present. All three sources are first-party (Anthropic) with no independent confirmation; the article attributes everything to the documentation, but does not say in the body that all sources are first-party (clarification filed). The user-profiles page itself carries no beta badge in the extracted text (only the webhooks page does), and the article does not call profiles beta, which is consistent. Sources array matches body URLs exactly (no orphans); the one internal link /article/2026-08/07-anthropic-launches-inference-hooks-... resolves to an existing published article.
Factual Accuracy: All numbers, names, headers, field names, endpoints and dates verified individually against the snapshots and live pages; no fabricated specifics. Title check: 'Can Target One End Customer Instead of a Whole Organization' is hedged ('can') and matches the docs ('a safety enforcement action ... can then apply to one end customer instead of your whole organization'); the article correctly explains that this requires attaching a profile and that without one an action applies to the whole organization. It does not overstate: restriction or ban is described as what Anthropic can place on the end customer a profile represents, and the article notes webhook delivery is not live and that enforcement fields exist only once enforcement actions are on for an organization (the latter is partly conveyed in What We Don't Know). Minor imprecision: the article says the Claude API, Claude Platform on AWS and Foundry deployments 'check the ID on every request'; the docs state Message Batches is an exception (batch creation checks only the format; per-request errors follow). Covered by a clarification.
Overall Assessment: Accurate, neutral, well-attributed News coverage of a first-party API change; only minor omissions, handled by two published clarifications. Approved with corrections. Featured candidacy: 1 - accurate first-party API release coverage for platform and reseller developers, narrow audience and no independent corroboration