Anthropic launches Inference Hooks, retires Claude Opus 4.1

Anthropic has introduced Inference Hooks in beta for Claude Enterprise, giving organizations a powerful new way to enforce governance on AI usage. The feature lets companies route governed prompts through an AI security server that performs allow or deny checks before the model processes them. That means enterprises can block unauthorized data from reaching Claude in real time, effectively building a DLP layer directly into the AI workflow. Crucially, it works across Claude's chat interface, Claude Code, and Cowork platforms, so enforcement is consistent regardless of how users interact with the model.

Alongside this release, Anthropic is announcing the retirement of Claude Opus 4.1, pushing users to upgrade to Claude Opus 5. While the older model is still available for a short time, Anthropic urges adoption of Opus 5 for better performance, safety, and alignment with current capabilities. For enterprises, the combination of Inference Hooks and the newer model strengthens both security and performance. The move reflects a broader industry trend toward AI safety tooling that sits outside the model itself — giving security teams full control over what enters the inference pipeline. As AI becomes more embedded in business processes, such pre-inference gates will likely become standard practice.