Policies - Abliteration.ai

Policies evaluate model traffic against questions you write. A policy can record a match without changing traffic, or block matching content. Create and manage policies in Console → Policies. Policies run on the standard /v1/chat/completions, /v1/responses, and /v1/messages endpoints. You do not need a separate API base URL.

How a policy works

Each policy has a yes-or-no question and four settings:

Setting	What it controls
Evaluates	Requests before model generation, responses after generation, or both. Response checks also cover tool calls.
Applies to	All projects, selected projects, or only requests where you explicitly select the policy.
Mode	Monitor records matches without interrupting traffic. Block stops a matching request or withholds a matching response.
Sensitivity	How readily content matches the question. Lower values flag more borderline content; higher values require a stronger signal.

You can have multiple policies affecting one request. Published policies scoped to a project apply automatically. Policies set to Only when selected apply when you add them to a Playground run or include their IDs in policy_ids on an API request. The platform’s baseline protections are separate from policies you configure.

Policy states

State	Effect
Draft	Not applied to traffic.
Monitoring	Published and active in Monitor mode.
Enforcing	Published and active in Block mode.
Paused	Temporarily inactive.

Publishing makes the current settings active. Saving edits as a draft does not change an existing published version. Pausing an active policy stops it from applying to new API requests.

Start here

Create a policy in Console.
Write a focused policy question and test it before blocking traffic.
Use policies in Playground or the API.
Review decisions to see matches and failed checks.

Last modified on September 30, 2026