Skip to content

Security and governanceBrief

Claude Tag API beta: what governance teams should check first

Anthropic released the Claude Tag API in beta on October 23 for Claude Enterprise organizations. It manages Claude Tag settings in Slack rather than calling a model, so the review centers on one Admin API key and what its scopes can change.

By INFRO Editorial

Published

5 min read

A grid of dark glass cards with small bar charts, a few of them lit amber, on a dark ground

In short

The Claude Tag API is a settings API: it reads and changes Claude Tag configuration on Slack channels and workspaces and attaches Access bundles 2. Review the Admin API key, the five settings it can change, and the Claude Tag data model it sits on top of, before you allow it.

Key takeaways

  • The Claude Tag API manages settings (enabled, instructions, model, environment, guest policy) and Access bundle attachments; it is not an inference endpoint.
  • The real exposure is the Admin API key: scopes are fixed at creation, the key belongs to the organization, and it keeps working after its creator leaves.
  • Start with a read:claude_tag key and add write:claude_tag only behind an approval workflow.
  • Claude Tag retains channel memory and session transcripts, so it is unavailable to ZDR and CMEK organizations; confirm your org's status first.
  • Ask Anthropic in writing whether API-made setting changes appear in your audit feed; the pages read do not say.

What the API is, and what it is not

Anthropic released the Claude Tag API in beta on October 23, 2026, for Claude Enterprise organizations, behind the experimental-claude-tag-2026-10-21 beta header 1. Claude Tag is Claude working as a teammate in Slack: someone mentions Claude in a channel and it answers in the thread 2. The API does not call a model. It reads and changes the Claude Tag settings an owner would otherwise set by hand in claude.ai 2.

It has 14 endpoints across four resources: Slack channels, Slack workspaces, Access bundles (read-only) and bundle targets, which attach a bundle to a channel or workspace 2. A request without the beta header returns 404, as if the endpoint did not exist 2. That matters for scanners: a 404 does not prove the API is off.

The documented responses return settings records, the source of each inherited value, and bundle attachments. They do not return message content. The data risk therefore comes from what the settings let Claude reach, not from the API's own payloads.

The five settings a write key can change

Only five settings are in scope: enabled, instructions, model, environment_id and guest_policy 2. Connecting workspaces, Slack defaults, creating bundles and spend limits stay in claude.ai 2. Each one maps to a review question:

SettingWhat a change can doAsk
enabledTurns Claude on in a channelWho approves a channel?
instructionsSets standing behavior for the scopeIs it reviewed like a prompt change?
modelPicks the modelIs the model on your approved list?
environment_idPicks the environment whose network access level governs egressWho may widen network access?
guest_policyrestrict, channel_only or allow (full access) in channels with guestsIs allow ever permitted?
settings the Claude Tag API exposes, from Anthropic's API page 2. The "ask" column is a review suggestion.

The egress link comes from Claude Tag's security page: outbound traffic from a channel sandbox is default-deny, and the environment's network access level is one of the allow layers 3. Attaching a bundle is the other lever. Bundles are Anthropic's Access bundles, and creating them is not in the API 2. The API can only attach existing ones.

Identity: the Admin API key is the whole story

Calls need an Admin API key with read:claude_tag for GET endpoints or write:claude_tag for POST and DELETE 2. The write scope does not grant reads 2. Only the primary owner creates the key, in claude.ai under Organization settings > API 2. Facts to put in the review:

  • Scopes are fixed at creation; adding one means a new key 4.
  • The secret is shown once, and the key prefix is sk-ant-api01- 4.
  • Keys belong to the organization and keep working after the creator leaves 2.
  • A primary owner can create a key that reaches every linked organization or one restricted to a single organization 4.

Choose the single-organization form. Anthropic's IP allowlist covers the Admin API and Compliance API only "with a key that covers only this organization" 5. A parent-wide key may fall outside that control; confirm with your account team. The allowlist is beta, takes up to 15 minutes to apply and holds up to 1,000 entries 5. Anthropic also advises short key expiration, since an allowlist limits where a key works, not for how long 5. The Enterprise key steps I read list a name and scopes only, so ask whether expiry is offered.

The Claude Tag runtime has its own identity model. In channels Claude acts under service accounts, not the requester's account, and a connector's credential is shared by everyone in the channels where it applies 3. Anthropic says each credential's use is logged 6. Review that separately from the API.

Data the feature sits on

Claude Tag retains channel memory and session transcripts, so it is not available to organizations with Zero Data Retention, a CMEK policy, or the HIPAA configuration on Claude Code and Cowork local mode 3. Check your organization's status first. If one of those applies, the question is moot.

Credentials you provision stay out of the sandbox. Agent Proxy injects them at request time and blocks requests that no allow layer covers 3. Setup screens are write-only 3. An allow-all * domain admits requests to any host on the listed ports, still without credentials 3.

Note two gaps. The IP allowlist does not cover Claude in Slack 5. And one-to-one DMs from members with a connected Claude account run on that member's own account, outside the organization spend limit 7. A policy written for channels does not describe DMs.

Evidence a reviewer will want

  1. Key register: name, scopes, owner, single-organization or parent-wide, creation date, rotation date, and where the secret is stored.
  2. A diff of Claude Tag settings per channel and workspace, taken from GET calls before and after rollout, with sources showing where each value comes from.
  3. Egress evidence: environment network level, allowed domains, and any * entries.
  4. Anthropic's security center report, if you use it. It compares settings with Anthropic's recommended configuration and exports as a report for auditors; it is beta guidance, not a certification 8.
  5. Change records for the key's callers. If an approval workflow turns channels on, keep the ticket ID beside each write.

The open question is audit coverage. The pages I read say the Compliance API Activity Feed records allowlist changes 5, but do not say whether Claude Tag settings changed through this API appear there. Get that answer in writing before you rely on it.

Keeping your own side of the workflow on record

The Claude Tag API is an organization-admin API in claude.ai, and Claude Tag work bills to your Anthropic usage balance under its own spend limit 7. A gateway in front of your applications' model calls will not see it. If your approval workflow runs on INFRO's management API or MCP server, INFRO's audit log (opens in a new tab) records member, key and policy changes with actor and timestamp 9. That covers INFRO's side of the workflow, not Anthropic's.

Questions readers ask

Does the Claude Tag API work for Claude Console (Claude Platform) organizations?
No. Anthropic's release notes and API page say it is for Claude Enterprise organizations only, and a key created in the Claude Console does not work with it 21.
Can the API connect a new Slack workspace or change spend limits?
No. Connecting a workspace, Slack defaults, creating or editing Access bundles, spend limits and every other Claude Tag setting beyond the five listed stay in claude.ai in this release 2.

From the team that builds it

Where this shows up in INFRO

INFRO is a control plane in front of the provider accounts you already hold. The capabilities below are the parts of it this article touches, each with the documentation that defines its behaviour.

INFRO accounts are self-serve. Provider usage is billed to you by your providers; INFRO charges a software subscription and takes 0% of model spend. See plans.

Sources

Numbered in order of first citation. Each was opened and checked against the claim on the date shown.

  1. 1.Claude Platform release notes (opens in a new tab) — Anthropic. Accessed Nov 6, 2026.
  2. 2.Claude Tag API (opens in a new tab) — Anthropic. Accessed Nov 6, 2026.
  3. 3.Security and data handling (Claude Tag) (opens in a new tab) — Anthropic. Accessed Nov 6, 2026.
  4. 4.Create an Admin API key (opens in a new tab) — Anthropic. Accessed Nov 6, 2026.
  5. 5.IP allowlisting (opens in a new tab) — Anthropic. Accessed Nov 6, 2026.
  6. 6.Claude Tag (opens in a new tab) — Anthropic. Accessed Nov 6, 2026.
  7. 7.Work with Claude Tag (opens in a new tab) — Anthropic. Accessed Nov 6, 2026.
  8. 8.More visibility and control over Claude Enterprise for security teams (opens in a new tab) — Anthropic. Accessed Nov 6, 2026.
  9. 9.Audit log (opens in a new tab) — INFRO. Accessed Nov 6, 2026.

About this article

Read on 2026-11-06: Anthropic's API release notes, the Claude Tag API page, the Admin API key and IP allowlist pages, the Claude Tag overview, security and product pages, Anthropic's November 5 Enterprise security post, and INFRO's audit log doc. Anything the pages did not state, such as whether API-made changes reach the Activity Feed or whether Enterprise keys can expire, is posed as a question, not asserted.

Last checked against its sources on . The next review is due by . No separate named reviewer is recorded for this article.

How articles are written, sourced and corrected is set out in the editorial standards.