Integrations · Model providersAnthropic, with sensitive data .
Your applications keep calling Anthropic the way they do today. RedactWall checks what they send and what comes back, and stops what should not leave your organization.
What that looks like
- A developer’s assistant pastes a customer file into a prompt → stopped before it reaches Anthropic, and the attempt is on the record.
- The model returns an account number in its answer → caught on the way back, before your application shows it.
- An analyst tries a workaround the policy forbids outright → refused, because 21 hard stops cannot be switched off by anyone.
What it takes to switch on
One setting per application: the address it already sends to. Your developers keep the library they use today, and there is no SDK to install.
How far this is proven
RedactWall has been run against Anthropic’s live service under a real account, on 2026-08-14. The request went through, and the decision on the way out and the decision on the way back were both recorded.