# AI compliance

AI compliance is showing that an AI system meets the rules that apply to it, whether laws, standards or your own internal policy. It takes written policies, enforcement at run time and evidence.

## Which rules apply

- **Laws.** The [EU AI Act](/glossary/eu-ai-act) for AI in the European Union, [HIPAA](/glossary/hipaa) for US health data, and the privacy and consumer laws of each place you serve.
- **Standards.** Reports that customers ask for, such as [SOC 2](/glossary/soc-2).
- **Internal policy.** What your company has decided its AI may write and do.

## What it takes

1. **Policies.** Each rule is written down as something a check can decide: what the AI may write, and what its agent may do.
2. **Enforcement.** A check runs on inputs, outputs and tool calls. A rule that lives only in a document or a system prompt is not enforced.
3. **Evidence.** Records show the rule was applied: what was asked, what the AI did, which policy fired. See [audit logs](/glossary/audit-logs).

Auditors ask for the evidence. A policy with no log behind it is only a claim.

Rules differ by country, so one global policy set rarely fits. See [region-specific AI policies](/glossary/regional-ai-policies).

## With White Circle

Policies are written in plain language, enforced on every input, output and tool call, and can differ by the user's location or jurisdiction. [Analyze](/products/analyze) has audit logs for compliance and incident review.

