Skip to main content
ATLASATLAS

Governance

Risk, ownership, AI accessibility, and policy simulation — with explicit honesty about enforcement state.

Total capabilities

33

High risk

7

AI-accessible

26

Metadata-only

33

Ownership gaps

0

Stale

1

Policy simulator

Pick a principal, capability, and action. ATLAS evaluates matching policies and returns the decision, reasons, and obligations. Every simulation is recorded as an audit event.

Run a simulation to see the decision, reasons, and obligations here.

Active policies

  • P1
    External models restricted from confidential contextdenysimulated

    External-model assistants may not access confidential or restricted capabilities without explicit approval.

    actions: invoke_read, invoke_write

    obligations: use_internal_model_only

  • P5
    Production destructive changes require approvalmanual approvalmetadata only

    Any destructive write in production requires manual approval.

    actions: invoke_write, manage

    obligations: record_justification, request_owner_approval

  • P10
    Read-only tools for approved assistantsallowsimulated

    Approved coding assistants (Cursor / Windsurf / Claude Code / Internal Agent) may invoke read-only MCP tools without manual approval.

    actions: discover, view_docs, invoke_read, simulate

  • P20
    Missing owner prevents activationmanual approvalmetadata only

    Capabilities without an owner team may not be invoked in production.

    actions: invoke_read, invoke_write

    obligations: require_owner_approval