# SOC 2 for AI products

SOC 2 is an independent audit report on how a service company protects customer data. It covers the company's security controls, not what an AI model writes or what its agent does.

## What it is

SOC 2 (System and Organization Controls 2) is a reporting framework from the AICPA (American Institute of Certified Public Accountants). An independent auditor examines a service company's controls against criteria for security and, if the company adds them, availability, processing integrity, confidentiality and privacy. The result is a report with the auditor's opinion, not a certificate.

There are two types:

- **Type I.** The controls are suitably designed, as of one date.
- **Type II.** The controls also operated effectively over a period of months.

## What it does not cover

A SOC 2 audit looks at how the company runs its systems: access, change management, encryption, incident response. It does not test what a model writes or what an agent does. A vendor with a clean report can still ship a model that follows a [prompt injection](/glossary/prompt-injection) or leaks data.

These are separate questions. SOC 2 is about the company. [Guardrails](/glossary/ai-guardrails), [red teaming](/glossary/ai-red-teaming) and [audit logs](/glossary/audit-logs) are about the model.

## With White Circle

We are SOC 2 Type II compliant (see the Trust Center). For what the audit does not cover, [Control](/products/control) enforces your policies on every input, output and tool call.

