> For the complete documentation index, see [llms.txt](https://help.nightfall.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://help.nightfall.ai/data-exfiltration-prevention/ai-agent-security/ai-governance/auditability-and-control/setup-and-installation/connect-inference-hooks.md). # Connect Inference Hooks Connect Nightfall as the Claude Enterprise Inference Hooks server. No endpoint agent and no MDM. Inference Hooks let Nightfall allow or deny a Claude Enterprise prompt before the model runs. Nothing is installed on the laptop, and there is no MDM rollout. This is not the device-hook install. Claude Code, Cursor, and VS Code hooks are separate guides on [Setup and Installation](https://help.nightfall.ai/data-exfiltration-prevention/ai-agent-security/ai-governance/auditability-and-control/setup-and-installation). Keep those when you need to stop a shell command or a tool call before it runs. ## Before you begin * A Claude Enterprise organization. Team, Pro, and Max cannot turn this on. * The Owner or Primary owner role in claude.ai. The Admin role cannot configure Inference hooks. * Inference Hooks enabled for your Nightfall tenant. If you do not see it on the Anthropic integration, contact your Nightfall account team. Inference Hooks and the Claude Compliance API are separate products on the same Anthropic integration. Connecting one does not connect the other. Keep the Compliance API connected. Claude records hook denials and circuit-breaker trips in the activity feed only when that API is enabled. ## Connect 1. In Nightfall, open the Anthropic integration and choose **Inference Hooks**. Copy the endpoint URL, the header name `X-Api-Key`, and the identification key. The key is shown once. The installation stays off until you paste the signing secret back. 2. In claude.ai, go to **Organization settings > Data and privacy > Inference hooks**. Turn on **Allow for your organization**. That does not start inspection. Choose **Configure**, paste the URL, add the `X-Api-Key` header, and save. Copy the signing secret Anthropic shows once. If you lose it, rotate it in the Claude admin and paste the new value. 3. Paste the signing secret into Nightfall. This step cannot be skipped. Nightfall shows a fingerprint of the secret it holds. Until the signature verifies, Claude allows every prompt. 4. Use **Test connection** in the Claude console. Re-enter the header value before you test. Nightfall should show that the test arrived and the signature verified. 5. Start in shadow mode: 100% inspected, timeout at least 5,000 ms, failure handling **Allow the request**. Then create the exfiltration policy. Existing device policies do not cover this traffic. See [Creating an AI Agent Security Policy](https://help.nightfall.ai/data-exfiltration-prevention/ai-agent-security/ai-governance/auditability-and-control/creating-an-ai-agent-security-policy). 6. When the decisions look right, set failure handling and turn on **Enforce verdicts**. **Allow the request** is the right default. **Block the request** stops Claude for the organization whenever Nightfall cannot answer in time. Changing the endpoint URL later clears every stored custom header. Re-enter `X-Api-Key` after any URL change. ## What the policy must select Inference Hooks evaluate **User Prompts** only. Tool Calls, Tool Responses, Shell Commands, and Model Responses do not fire on this integration. Leave MCP server scope on **All MCP servers**. Turn **content scanning on** and add detection rules when you want to block only prompts that contain sensitive data. {% hint style="warning" %} Content scanning off is live. It does not wait for a detector. **Block** denies every prompt the policy covers. **Monitor** records a violation on every prompt and lets it through. That is the same rule as other endpoint exfiltration policies. {% endhint %} ## If Nightfall cannot answer Claude applies the failure mode you set. Sustained failures trip Anthropic's circuit breaker. With **Block the request** selected, the organization is blocked until the breaker resets. Fix the cause, then turn **Enforce verdicts** back on. The trip is recorded in Claude's activity feed when the Compliance API is enabled. ## What a denied person sees The prompt does not reach the model. Claude shows Nightfall's reason, then the standing message under **Custom blocked prompt message**. The denied message stays in the conversation until it is edited or removed. On claude.ai, edit it or start a new chat. On Claude Code, run `/rewind` or `/clear`. On Cowork, edit the latest message or start a new task. In Slack, edit or delete the message, then send `@Claude !restart`. Nightfall does not receive raw file bytes or images. A screenshot is not inspected. Voice mode is not covered. Claude on Bedrock, Vertex, or a Console API key is not covered. Policy templates are on [Hook policy recommendations](https://help.nightfall.ai/data-exfiltration-prevention/ai-agent-security/ai-governance/hook-policy-recommendations). --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation by asking a question. Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters: ``` GET https://help.nightfall.ai/data-exfiltration-prevention/ai-agent-security/ai-governance/auditability-and-control/setup-and-installation/connect-inference-hooks.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.