inference-hooks-configuration changedmanage-claude/inference-hooks-configuration
Nearest release: v2.1.307, published an hour after this site recorded the change. Shown because the two are within 24 hours of each other. Nothing here says the release caused the edit.
Recorded here
Lines+13added
Lines−13removed
From line
10
where the diff opens
First seen
14 Aug 2026
this site's first read of the page
Recorded edits15to this page, all time
## Before you start ## Before you begin
The whole hunk
from line 10, old and new numbered
/
from line 10
1010
1111Inference hooks send prompts from your organization to an AI security server you choose, and hold each request for an allow or deny verdict before Claude processes it. This page walks through turning the feature on, connecting your server, and controlling enforcement. To learn what Inference hooks are and when to use them, see the [Inference hooks overview](https://platform.claude.com/docs/en/manage-claude/inference-hooks). To build the AI security server itself, see [Develop an Inference hooks integration](https://platform.claude.com/docs/en/manage-claude/inference-hooks-endpoint).
1212
13## Before you begin
13## Before you start
1414
1515You need:
1616
from line 29
2929 </Step>
3030
3131 <Step title="Open the Inference hooks settings page">
32 Still in **Data and privacy**, open the **Inference hooks** section to reach the Inference hooks settings page. It lives under Data and privacy rather than as its own entry in the settings nav, so its breadcrumb reads **Data and privacy / Inference hooks**. Until you save an endpoint, the page warns that prompts aren't being inspected yet, and **Enforce verdicts** stays off with a **Requires endpoint** badge.
32 Still in **Data and privacy**, open the **Inference hooks** section to reach the Inference hooks settings page. It lives under **Data and privacy** rather than as its own entry in the settings nav, so its breadcrumb reads **Data and privacy / Inference hooks**. Until you save an endpoint, the page warns that prompts aren't being inspected yet, and **Enforce verdicts** stays off with a **Requires endpoint** badge.
3333 </Step>
3434
3535 <Step title="Configure your endpoint">
36 Click **Configure** to open the **Set up endpoint** dialog and enter the **Endpoint URL**: the `https://` URL that receives verdict requests. Only `https://` URLs are accepted.
36 Click **Configure** to open the **Set up endpoint** dialog box and enter the **Endpoint URL**: the `https://` URL that receives verdict requests. Only `https://` URLs are accepted.
3737
38 The dialog asks for nothing else at this point: custom request headers come in step 5, and failure handling in step 6. Click **Next** to save. Once an endpoint is saved, the button reads **Edit**.
38 The dialog box asks for nothing else at this point: custom request headers come in step 5, and failure handling in step 6. Click **Next** to save. Once an endpoint is saved, the button reads **Edit**.
3939 </Step>
4040
4141 <Step title="Store your signing secret">
from line 45
4545 </Step>
4646
4747 <Step title="Add request headers and test the connection">
48 Clicking **Next** on the signing secret dialog reopens the endpoint dialog, now with two more controls:
48 Clicking **Next** on the signing secret dialog box reopens the endpoint dialog box, now with two more controls:
4949
5050 * **Custom request headers:** up to 16 static headers sent with every verdict request so your AI security server can authenticate the caller. Header values are stored encrypted and never shown again; after saving, only the header names are displayed. Because values are write-only, saving any change to the headers requires re-entering every value. Changing the endpoint URL clears all stored header values so your credentials are never sent to a new destination; re-enter them after a URL change. Header names must use standard HTTP token characters with `-` rather than `_`, and must not collide with reserved names (request-framing headers such as `Content-*` and `Host`, proxy and cookie headers, client-address headers such as `X-Forwarded-*`, the `webhook-*` signature headers, and the `X-Anthropic-*` prefix). Values must be printable ASCII.
5151 * **Test connection:** Claude sends a synthetic test prompt to the URL and headers currently in the form, not the saved values, so re-enter any stored header values before testing. On success, the result reports whether your AI security server returned an allow or a deny verdict for the test prompt, which surfaces a deny-everything default before you start enforcing. It also shows the round-trip time Anthropic measured for that one test request, on a new connection, including the connection handshake.
from line 66
6666 </Step>
6767
6868 <Step title="Choose failure handling and timeout">
69 Under **Failure handling**, set **Mode** to choose what happens while the AI security server is unreachable or verdicts time out:
69 Under **Failure handling**, set **Mode** to choose what occurs while the AI security server is unreachable or verdicts time out:
7070
7171 * **Block the request:** stop inference when your AI security server can't deliver a verdict (fail closed).
7272 * **Allow the request:** let the request proceed to the model without inspection (fail open).
7373
74 The dropdown's third option, **Shadow mode**, is a rollout tool rather than a failure policy; see [Shadow mode](https://platform.claude.com/docs/en/manage-claude/inference-hooks-configuration#shadow-mode).
74 The dropdown menu's third option, **Shadow mode**, is a rollout tool rather than a failure policy; see [Shadow mode](https://platform.claude.com/docs/en/manage-claude/inference-hooks-configuration#shadow-mode).
7575
7676 Then set **Prompt verdict timeout (ms)**: 1 to 10,000ms, with a default of 5,000ms. The budget covers the entire exchange, and a slower verdict counts as an unreachable server, so set the lowest value your server can reliably meet.
7777
from line 87
8787 <Step title="Turn on Enforce verdicts">
8888 To evaluate verdicts against live traffic without blocking anyone at first, set **Mode** to **Shadow mode** (step 6) before turning on enforcement; see [Shadow mode](https://platform.claude.com/docs/en/manage-claude/inference-hooks-configuration#shadow-mode).
8989
90 Turn on **Enforce verdicts** to gate Claude on your AI security server's verdict for every governed prompt, then confirm in the dialog, which restates your failure handling choice. Allow about a minute for the change to reach every Anthropic server; requests already in flight finish under the old setting. Turning it off stops prompts from being sent to your AI security server, again within about a minute; your configuration is kept.
90 Turn on **Enforce verdicts** to gate Claude on your AI security server's verdict for every governed prompt, then confirm in the dialog box, which restates your failure handling choice. Allow about a minute for the change to reach every Anthropic server; requests already in flight complete under the old setting. Turning it off stops prompts from being sent to your AI security server, again within about a minute; your configuration is kept.
9191
9292 **Validate tool calls**, below **Enforce verdicts**, also sends the tool calls in each of Claude's responses to your AI security server and waits for its verdict before they run; see [The tool call frame](https://platform.claude.com/docs/en/manage-claude/inference-hooks-endpoint#the-tool-call-frame). It is on by default in a new configuration. It has no effect while **Enforce verdicts** is off, and a change to it takes about a minute to reach every Anthropic server, as with **Enforce verdicts**.
9393 </Step>
from line 101
101101
102102## Exclusions
103103
104Under **Exclusions**, select roles whose members are not covered by Inference hooks: their prompts are never sent to your AI security server. Only custom roles your organization created can be excluded; the built-in roles aren't offered. Pick them in the role selector, whose placeholder reads **Select roles to exclude**, and manage who holds each role from the roles admin page (**Manage roles**); changing exclusions requires identity management permission. The list is empty by default, and with no roles excluded, every governed request is inspected.
104Under **Exclusions**, select roles whose members are not covered by Inference hooks: their prompts are never sent to your AI security server. Only custom roles your organization created can be excluded; the built-in roles aren't offered. Select them in the role selector, whose placeholder reads **Select roles to exclude**, and manage who holds each role from the roles admin page (**Manage roles**); changing exclusions requires identity management permission. The list is empty by default, and with no roles excluded, every governed request is inspected.
105105
106106Exclusion applies to a user's interactive sessions. Changes to the exclusion list are recorded in your organization's [Activity Feed](https://platform.claude.com/docs/en/manage-claude/compliance-activity-feed) as role permission changes (`rbac_role_permission_added` and `rbac_role_permission_removed`).
107107
from line 114
114114The endpoint health area of the Inference hooks settings page shows:
115115
116116* **Endpoint status:** Healthy, Tripped, Not enforcing, or Not configured before an endpoint is saved.
117* **Failures per minute:** webhook failures over the last two minutes, averaged.
117* **Failures per minute:** webhook failures over the last 2 minutes, averaged.
118118* **Block rate:** denials as a share of your AI security server's verdicts, shown while the rollout percentage is below 100.
119119* **Circuit breaker tripped:** when the breaker last tripped, if it has.
120* **Recent errors:** each entry shows when the failure happened, an error type, and a category. The category is `webhook_error` for a problem with your endpoint or the connection to it, or `relay_error` for a failure inside Anthropic's systems. Entries never include request content or your endpoint URL. The list keeps the 10 most recent failures and clears an hour after the last one.
120* **Recent errors:** each entry shows when the failure occurred, an error type, and a category. The category is `webhook_error` for a problem with your endpoint or the connection to it, or `relay_error` for a failure inside Anthropic's systems. Entries never include request content or your endpoint URL. The list keeps the 10 most recent failures and clears an hour after the last one.
121121
122122The error types in **Recent errors** mean:
123123
from line 139
139139
140140Sustained webhook failures attributable to your AI security server trip the circuit breaker, which stops enforcement: your server is no longer contacted, and your **Failure handling** choice applies to every inspected request. With **Block the request** selected, users in your organization are blocked until the breaker resets. When the breaker trips, administrators are also notified in the claude.ai notification center.
141141
142Each trip is also recorded in your organization's [Activity Feed](https://platform.claude.com/docs/en/manage-claude/compliance-activity-feed) as an `inference_hooks_circuit_breaker_tripped` activity, so your security team or vendor can alert on trips from monitoring they already run, such as a SIEM that ingests the feed. One activity is recorded per trip, not one per affected request. Recording requires the Compliance API to be enabled for your organization; see [Set up the Compliance API](https://platform.claude.com/docs/en/manage-claude/compliance-api-access).
142Each trip is also recorded in your organization's [Activity Feed](https://platform.claude.com/docs/en/manage-claude/compliance-activity-feed) as an `inference_hooks_circuit_breaker_tripped` activity, so your security team or vendor can alert on trips from monitoring they already run, such as a SIEM that ingests the feed. One activity is recorded for each trip, not one for each affected request. Recording requires the Compliance API to be enabled for your organization; see [Set up the Compliance API](https://platform.claude.com/docs/en/manage-claude/compliance-api-access).
143143
144144To recover, fix the server, then turn **Enforce verdicts** back on to reset the breaker.
145145
from line 161
161161
162162There are two levels of off:
163163
164* **Enforce verdicts** off, on the Inference hooks settings page: within about a minute, prompts from your organization stop being sent to your AI security server; requests already in flight finish under the old setting. The settings page stays available, so use this to pause enforcement while you work on your AI security server.
164* **Enforce verdicts** off, on the Inference hooks settings page: within about a minute, prompts from your organization stop being sent to your AI security server; requests already in flight complete under the old setting. The settings page stays available, so use this to pause enforcement while you work on your AI security server.
165165* **Allow for your organization** off, in **Data and privacy** settings: prompts are no longer inspected, and the Inference hooks settings become unavailable until you turn it back on. Your endpoint configuration, custom headers, and signing secret are kept either way; turning it back on forces **Enforce verdicts** off and clears a tripped circuit breaker, so turn enforcement on again when you are ready.
166166
167167## Next steps
No line in this hunk matches that.