How policies work
When a user asks a question, Human Intelligence parses the user’s intent and checks it against active policies before any data is queried. If a policy restricts the request, the user receives a clear explanation of the boundary, ensuring agent interactions remain compliant with the policies set by your organization.Examples
Policies are defined using flexible, plain-text narrative guidance. Here are a few common patterns:- Personnel Recommendations: “Never make explicit recommendations or suggestions for terminations, layoffs, or hiring decisions.” (The agent can show performance data, but cannot tell a manager who to fire).
- Legal & Compliance Guardrails: “Do not attempt to provide legal compliance opinions on wage/hour exemptions or independent contractor classifications.”
- Culture and Bias Restraints: “Never generate forced stack-rankings of employees within a department based on qualitative notes.”
Policies are available on the Enterprise tier. If you’re on a different plan, categories and roles handle your access control.
Policies and AI assistants
When you connect an AI assistant to Human Intelligence, narrative policies are enforced directly at query time. Because the policy engine evaluates the intent of the request, the assistant will block or gate restricted behavior even if a user phrases their question creatively or attempts a prompt-injection attack. This ensures your organization’s ethical and legal guardrails travel with your people data, no matter how it is accessed.Monitoring
Every time a Policy is triggered, an entry is written to the Audit Log with immutable records for compliance tracking. The audit log captures:- Timestamp & User ID: Who made the request.
- The Raw Input Prompt: The exact text submitted.
- Policy Triggered: Which specific policy triggered for this request.
- Action Taken: Whether the request was blocked, modified, or flagged.
Want to learn more about policies or upgrade to Enterprise? Contact our team and we’ll walk you through it.