> This page is for Tools.

> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.moveworks.com/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.moveworks.com/_mcp/server.

# Policy Validators

## What are Policy Validators?

Natural language rules are probabilistic. "Don't show contractor benefits to FTEs" as a system prompt works most of the time. But LLMs can misinterpret rules, users can jailbreak them, and the more policies you add, the less likely any single one gets enforced ([Lost in the Middle](https://arxiv.org/abs/2307.03172) problem). Your HR legal team doesn't accept 90% compliance. Policy Validators guarantee 100%.

While it's appropriate to steer some behavior through natural language, enforcing **business rules** through natural language can be unreliable.

* Natural language policies are useful when there is a lot of nuance, but they are too imprecise for strict rules.

  <table>
    <thead>
      <tr>
        <th>
          Example
        </th>

        <th>
          Optimal Steering Behavior
        </th>
      </tr>
    </thead>

    <tbody>
      <tr>
        <td>
          "If the escalation seems important, surface it to the on-call engineer"
        </td>

        <td>
          **Natural Language**

          Let the LLM decide if it's important.
        </td>
      </tr>

      <tr>
        <td>
          "If the employee is not an FTE, don't show them our FTE benefits"
        </td>

        <td>
          **Rules**

          ```
          employee.status IN ["Contractor", "Contingent"]
          ```
        </td>
      </tr>
    </tbody>
  </table>
* Natural language policies can be misunderstood by LLMs, or users might jail break them.
* The more policies you add, the less likely your AI agent will reliably enforce policies ([Lost in the Middle](https://arxiv.org/abs/2307.03172) problem).

Policy Validators enforce & guarantee that rules are **always** respected accurately, no matter how many you add.

Learn more about [our approach to policy validators here](https://www.moveworks.com/us/en/resources/blog/how-policy-validators-help-ai-compliance).

## How do you configure Policy Validators?

We have policies embedded in different parts of our product to provide you control where you need it

* **[Activity Confirmation Policies](/tool-studio/conversation-process/activities#activity-confirmation-policies)** - enforces user approval before an action in taken.
* **[Slot Validation Policies](/tool-studio/conversation-process/slots#slot-validation-policy)** - validates input values satisfy constraints
* **[Slot Inference Policies](/tool-studio/conversation-process/slots#slot-inference-policy)** - decides if the LLM must ask the user for an input.
* **[Conversational Process Decision Policies](/tool-studio/conversation-process/control-flow#conversation-process-decision-policy)** - controls agentic decision-making with rules

\