Core Concepts

The Responsibility Layer

SOVR sits between your AI agent and the actions it wants to execute.

Default Deny

Unlike traditional systems that allow by default, SOVR operates on zero trust:

• Every action must pass through the gate
• Unknown actions are denied by default
• Explicit policies define what's allowed

Risk Assessment

Each action is scored based on:

Factor	Weight	Description
Action Type	30%	Category risk (financial, data, communication)
Context	25%	Recipients, amounts, sensitivity
History	20%	Past decisions for similar actions
Anomaly	15%	Deviation from normal patterns
Time	10%	Business hours, frequency

Decision Types

Allow

Action is safe to execute immediately.

Require Approval

Action needs human sign-off before execution.

Deny

Action is blocked with explanation.

Audit Chain

Every decision creates a tamper-proof record:

{
  "id": "aud_123",
  "timestamp": "2026-02-05T05:00:00Z",
  "action": "send_email",
  "decision": "allow",
  "riskScore": 15,
  "previousHash": "sha256:abc...",
  "hash": "sha256:def..."
}

Trust Bundles

Export cryptographically signed evidence packages for:

• Compliance audits
• Legal discovery
• Insurance claims
• Customer assurance