Metomic adds Claude Enterprise prompt screening for sensitive data

Metomic now inspects Claude Enterprise prompts before Anthropic’s model sees them, using customer-defined data policies to allow or block requests in real time. The integration arrives as Anthropic’s inference hooks move from beta into enterprise workflows and gives security teams a faster way to control sensitive data in AI use.

Why it matters: - Claude Enterprise customers can stop sensitive data from reaching Anthropic’s model before inference runs. - The integration gives security teams a policy checkpoint for employee prompts without adding device agents or new infrastructure. - The move addresses a common enterprise AI problem: controlling data exposure without slowing adoption.

What happened: - Metomic added support for Anthropic’s Claude inference hooks to its AI data security platform. - Claude Enterprise customers can route employee prompts through Metomic, which returns an allow or deny verdict before the model runs. - Anthropic released inference hooks in beta on Aug. 5. - Anthropic’s hook covers claude.ai, Claude Code and Claude Cowork.

The details: - Customers define what counts as sensitive data, and Metomic enforces those policies at the prompt layer. - Metomic’s inspection engine already classifies data across Slack, Google Workspace, Salesforce and other SaaS tools. - The integration works as an inspection layer behind Anthropic’s routing system. - Metomic says the setup can go live in under 30 minutes. - Teams start in observation mode, where Metomic records and analyzes traffic without blocking requests. - Teams can then switch to enforcement after establishing a baseline. - Custom sensitive data labels are in early access. - Those labels let security and compliance teams define sensitive context in natural language instead of keyword lists.

Between the lines: - Anthropic built the routing, but customers keep control over policy decisions. - Metomic is positioning semantic detection as a better fit for enterprise data than keyword filtering. - Bethany Ayers, Metomic’s CEO, said keyword lists can block legitimate work, encourage employees to route around security and create controls people do not trust. - The pitch is that context-aware detection is the hard part, and that is where Metomic is aiming to differentiate itself.

What's next: - Claude Enterprise customers can use Metomic’s Claude inference hook support now. - Security teams can test in observation mode first, then move to blocking once policies are tuned. - Wider adoption will likely depend on how well the natural-language policy labels handle real enterprise data. - Metomic will continue to pair its Claude support with its broader AI data security platform.

The bottom line: - Metomic is turning Anthropic’s new inference hooks into a real-time policy gate for enterprise AI prompts, with customer-defined data controls deciding what reaches Claude.