Implementation Guide
Adding an AI System to an Existing SOC 2 Scope
Treat the AI system as a change to your existing control environment, not a separate program. Define which components belong inside the boundary, update the system description and risk assessment, map those components to the criteria already in scope, extend existing controls and close gaps, then agree with your auditor on timing, model provider treatment, and evidence before the change goes live.
Four steps from AI deployment to audit evidence
The work moves in sequence: boundary, documentation, controls, then evidence and timing. Each step produces something the next one depends on.
| Step | Focus | Documented output |
|---|---|---|
| 1. Define the boundary | Models, prompts, retrieval sources, tools, identities, and providers. | Component inventory, data flow, and boundary decisions with rationale. |
| 2. Update description and risk | Reflect AI components, data flows, and AI-specific risks. | Updated system description and risk assessment. |
| 3. Map and extend controls | Apply existing criteria and controls, then record gaps. | Gap review of access, change, logging, vendor, and incident controls. |
| 4. Plan evidence and timing | Agree on period treatment and collection with the auditor. | Control start dates and an evidence plan for each control. |
Start by defining the AI system boundary
Decide which components belong inside the boundary. The components to consider are models, prompts, retrieval sources, tools, identities, and providers.
- Inventory AI components and draw the data flow, including external providers.
- Record the boundary decision and rationale for any excluded component.
Update the system description and risk assessment
Once the boundary is set, reflect it in the documents your auditor relies on. Add AI features, AI components, data flows, and AI-specific risks to the system description and risk assessment.
Identify gaps and plan evidence
Map AI components to controls you already operate
Map the AI components to the criteria already in scope, then extend existing controls and close gaps. Apply existing criteria and controls first, and record the gaps that remain.
Run a gap review of the following control areas against the AI components:
- Access
- Change
- Logging
- Vendor
- Incident
Plan how each control produces evidence
Document control start dates and confirm how each control will produce evidence. Agree on evidence collection with your auditor before the change goes live.
Decide whether to wait or communicate mid-period
Agree with your auditor on timing, model provider treatment, and evidence before the change goes live. This includes how the audit period treats the change.
- Classify model providers and agree on carve-out or inclusive treatment with your auditor.
- Agree on period treatment and collection with the auditor.
Implementation checklist
- Inventory AI components and draw the data flow, including external providers.
- Record the boundary decision and rationale for any excluded component.
- Add AI features and AI-specific risks to the system description and risk assessment.
- Classify model providers and agree on carve-out or inclusive treatment with your auditor.
- Run a gap review of access, change, logging, vendor, and incident controls against the AI components.
- Document control start dates and confirm how each control will produce evidence.
Plan evidence before the audit period asks for it
If you are extending SOC 2 controls to AI agents, talk with Trussed AI about how runtime governance, agent identity, and audit logging can support your evidence planning.
Talk to an Expert