Anthropic launches inference hooks for Claude Enterprise security

Bridging AI and corporate security

Anthropic recently announced new security features for its enterprise customers, specifically making inference hooks available for Claude Enterprise. These hooks act as interfaces that allow organizations to redirect user prompts and tool calls to their own internal systems. The primary goal is to prevent data loss while allowing companies to keep using generative AI models.

By using these hooks, a business can ensure that every interaction with the AI follows the same rules as their other corporate software. This means that established security policies do not have to be bypassed or weakened to accommodate new technology. It provides a way for IT departments to maintain oversight over what information is being shared with the model.

As BornCity reported, these hooks are designed to work without breaking the existing system architecture. Organizations can plug in their own verification tools to scan inputs before they ever reach the AI. This creates a layer of protection that sits between the user and the model, ensuring that sensitive data is filtered out automatically.

How the hooks function

The new system is divided into two specific types of hooks: Prompt-Hooks and Pre-Tool-Call-Hooks. Prompt-Hooks intercept the initial text or instructions a user types into the system. Pre-Tool-Call-Hooks work similarly but focus on the actions the AI might take, such as using a specific software tool or performing a task.

When a user submits a request, the system pauses the action to let it pass through a third-party checking tool. Once the check is complete, the request proceeds to the AI. This setup allows for real-time monitoring and filtering of both the information going in and the actions being requested by the model.

The system includes specific technical limits to ensure performance remains stable. For example, there is a standard timeout of 5 seconds, but administrators can configure this between 1 and 10,000 milliseconds. To maintain high security, the system blocks public tunnel services like ngrok, which could otherwise be used to bypass corporate network rules.

Broad coverage across the ecosystem

One of the key advantages of this update is that a single hook configuration can protect multiple different services at once. This includes the main claude.ai web interface and the Cowork platform. It also extends to Claude Code, which is used for programming, and various integrations for Microsoft 365.

Furthermore, the security hooks apply to Claude Tag and various Chrome extensions. This wide range of coverage ensures that employees cannot easily find a

Advanced administrative controls

Administrators can manage multiple Claude organizations through a centralized Parent Console located at admin.claude.com. This console allows for high-level management of single sign-on, domain verifications, and automated user provisioning via SCIM. However, users must be aware that setting up this overarching console is a permanent action that cannot be undone once it is established.

To further restrict network access, the system includes an IP allowlist. This feature can store up to 1,000 entries in IPv4, IPv6, or CIDR formats. Additionally, a new remote logout function via the administration API allows managers to end active user sessions across all browsers and devices. This specific function is designed to terminate active sessions while preserving API keys, running cloud sessions, and scheduled routines.

Reporting and regulatory guidance

Anthropic is also introducing tools to help businesses manage the documentation of their security posture. The Security Center includes a feature that allows for the export of reports, which can include confirmations as a PDF file. This helps administrators document the risks they have accepted and maintain a record of their security configurations for audit purposes.

Because new regulations like the EU AI Act are becoming active, companies need clear paths for compliance. Anthropic provides a free implementation guide that offers a compact overview of the requirements, deadlines, and risk classes of the current AI regulation. This resource is intended to help IT and legal departments stay up to date with the evolving legal landscape of artificial intelligence.