# Legal and Compliance Guide

*Audience: legal, compliance, and audit teams. This guide has no UI steps - for those, see the [Admin Configuration Guide](/docs/ai-use-cases/admin-configuration-guide/).*

:::info
This guide describes how AI use case records can support your AI governance program. It isn't legal advice, and using Relyance doesn't by itself make your organization compliant with any law or standard. Confirm your obligations with your own counsel.
:::

## Why govern AI at the use case level

AI regulations and standards attach obligations to an AI *system* deployed for a purpose, not to a model on its own. The same model can be low-risk in one product and high-risk in another, depending on what it decides, about whom, and with what data. An AI use case in Relyance is that system: a business purpose, an accountable owner, and the models, agents, MCP servers, and vendors it runs on.

Because components are shared across use cases, a change to one model or agent shows up on every use case that relies on it - so a finding about a shared component doesn't stay hidden inside one team's review.

## The obligations a use case record supports

| Expectation | How an AI use case supports it |
|---|---|
| **An inventory of AI systems** - knowing what AI you run, where, and who owns it (for example, the *Map* function of the NIST AI RMF and the AI system inventory expected by ISO/IEC 42001) | Each use case records its purpose, department, business owner, and review owners, and lists the discovered components it's built from |
| **Risk classification** - deciding how risky each system is and applying controls in proportion (for example, the EU AI Act's risk tiers) | Each use case carries a risk level, and the AI Governance and Risk Assessment includes an AI system classification section and additional sections for high-risk AI systems |
| **Impact and risk assessment before deployment** | The intake triages each use case, and the AI Governance and Risk Assessment covers objective and proportionality, data governance, transparency, human oversight, and recordkeeping. Completed assessments can raise risks into the Risk Register, linked back to the use case |
| **Data governance** - knowing what personal and sensitive data an AI system processes | Data use and sensitive data exposure are derived from what Relyance observes in the components, rather than from questionnaire answers alone |
| **Ongoing monitoring and change control** - re-assessing when a system changes materially | Where change management is enabled, an approved use case is compared with every scan; material changes are scored, held for review, and reopen the affected assessment questions. Review dates are scheduled from the risk level |
| **Documentation and recordkeeping** | Every create, edit, component change, approval, change decision, and lifecycle move is recorded in the use case's audit log, and the compliance package bundles the evidence for one system |

## What the records demonstrate

- **That the system is known and owned** - a named business owner, review owners, department, and a description of its purpose.
- **What it's built from** - the specific models, agents, MCP servers, and vendors, discovered from your environment rather than self-reported, and whether each is registered.
- **That it was assessed** - the intake and risk assessment, their answers, and who completed them, with risks raised from them recorded in the Risk Register.
- **How it's monitored** - the policies that apply to its components, their status, and the issues raised against them.
- **That changes were controlled** - what was approved and when, every change since, how each was scored, and who or what accepted it and why. Decisions the governance engine made are labelled as the engine's, so they can be told apart from a colleague's.

## Limits to be aware of

- **Discovery depends on your integrations.** A use case can only show components, data use, and changes that Relyance can observe through the sources you've connected.
- **Vendors don't count toward governance status.** A use case is **Registered** only when every attached model, agent, and MCP server is registered; vendors are listed but don't affect the rollup.
- **The change score is deliberately simple.** It's a deterministic points-based score designed to route changes to the right reviewer, not a legal determination of materiality. Your reviewers remain responsible for that judgment.
- **Quarantine is a record, not a kill switch.** Quarantining a use case flags it in Relyance; it doesn't stop the AI system in your environment.

## Related

- [Audit Log and Reporting](/docs/ai-use-cases/audit-log-and-reporting/) - what's recorded and how to export it.
- [Assessments: Legal and Compliance Guide](/docs/assessments/legal-and-compliance-guide/).
- [Risk Register: Legal and Compliance Guide](/docs/risk-register/legal-and-compliance-guide/).
