Skip to main content
Skip table of contents

Governance workflows by role

This page describes how different roles interact with Saidot and how governance work flows in practice — both through the UI and programmatically via the REST API or MCP servers. It also documents the pre-built workflow Skills available to teams who want a ready-made starting point.

This page assumes familiarity with Saidot's data model. If you have not yet read How Saidot governance works, start there — it explains what Systems, Risks, Controls, the Library, and the Governance space are.

Pre-built workflow Skills

Saidot provides a set of open-source, pre-built workflow automations called Skills. These are ready-to-use agentic workflows that an AI assistant (such as Claude) can carry out on your behalf when connected to Saidot's MCP servers.

Preview: Skills are currently in preview and delivered to customers on request. They are functional and actively used, but the list, names, and behaviour may change as the platform evolves.

Important: Skills are starting points, not finished solutions. Every organisation has its own governance processes, terminology, guardrails, and approval chains. Review each Skill before using it and adapt it to match your organisation's needs and practices.

Examples of typical skills:

  • library — Read-only exploration of Saidot's curated Library: models, products, risks, controls, policies, and tasks. Use to research content before registering or governing assets.

  • system-registration — Register and onboard a new AI system, collect metadata, and link underlying assets (models, datasets)

  • asset-registration — Register models, agents, and products as governance assets independently of any system

  • data-governance — Document and govern datasets, profile data sources, link to systems, and flag privacy risks

  • risk-management — Run a full risk analysis on a system: triage inherited risks, find coverage gaps, recommend Library risks, write evaluations and treatment plans

  • eu-aia-prep — Readiness check for EU AI Act classification: verify required fields are populated and prepare for the conversational classifier

  • transparency-reporting — Generate a transparency report (PDF) and a live KPI snapshot for a system

  • gov-report — Generate a self-contained HTML governance report with risk, control, and system profile data

How Skills are triggered: Just describe what you want in the conversation — the agent will select the appropriate Skill based on what you say. You do not need to name a Skill explicitly or use any special syntax. If you want to direct the agent to a specific Skill, you can mention it by name, but this is optional, not a requirement.

AI governance lead

Governance leads are responsible for the overall AI governance programme — maintaining the AI inventory, tracking risk posture, and ensuring compliance deadlines are met.

Typical actions:

  • Query the full AI system inventory across spaces

  • Check aggregate risk posture across all systems, filtered by status and treatment

  • Identify systems with no owner or incomplete governance

  • Generate transparency reports for external stakeholders

  • Trigger EU AI Act classification readiness checks

  • Produce a governance dashboard for leadership

Recommended MCP profile: Governance MCP (read + limited write), Docs MCP (read)

Risk and compliance team

Risk and compliance teams own the risk register, evaluate control effectiveness, and map governance to regulatory frameworks.

Typical actions:

  • Identify risks without controls or with overdue reviews

  • Import Library risks onto a system

  • Bulk-add controls from the Library to address a risk gap

  • Request a risk review from a subject matter expert

  • Run a full risk gap analysis

  • Map controls to regulatory requirements

Recommended MCP profile: Governance MCP (read/write), Library MCP (read), Docs MCP (read)

System owner

System owners are accountable for an individual AI system — keeping its governance record current, confirming risk treatments, and signing off on lifecycle transitions.

Typical actions:

  • Get a full picture of a specific system's current state

  • Review and update risk treatment plans

  • Confirm control implementation status

  • Update system lifecycle stage on deployment or decommission

  • Link a newly deployed model to the system

  • Generate a transparency report for a specific system

Recommended MCP profile: Governance MCP (read/write scoped to their systems)

Developer and platform engineer

Developers integrate Saidot into CI/CD pipelines, deployment workflows, and internal tooling. They are typically the ones who configure MCP servers and build automation that calls the Governance API.

Typical actions:

  • Register a new AI system at deploy time

  • Register a new model version and link it to systems

  • Document a dataset used for fine-tuning or RAG

  • Ingest observability events from production (model drift, incident signals)

  • Check approval status of a model before deployment

  • Set up event-driven governance triggers: see Event-driven governance patterns

Recommended MCP profile: Governance MCP (read/write), Docs MCP (read)

Reviewer

Reviewers are subject matter experts — legal, security, ethics — who are assigned to review specific risks, controls, or systems on request.

Typical actions:

  • List open review requests assigned to you

  • Get the full context of a system before reviewing

  • Submit a control or risk review decision

  • Add a comment or observation to a risk

Recommended MCP profile: Governance MCP (read + targeted write for review actions), Docs MCP (read)

Choosing what to automate

Not all governance actions benefit equally from automation. A useful heuristic:

  • High-value automation targets: Repetitive data entry (registering systems from descriptions), bulk operations (adding standard controls across a portfolio), scheduled reporting, onboarding checklists, and any action triggered reliably by an external event (deployment, incident)

  • Keep human in the loop: Risk treatment decisions, classification and risk level assignments, review sign-offs, and any action with regulatory or reputational consequences

  • Agent-assisted, human-approved: Draft risk evaluations, suggest controls from the Library, pre-fill system metadata — then route for human review before saving

JavaScript errors detected

Please note, these errors can depend on your browser setup.

If this problem persists, please contact our support.