Authority Boundaries

Authority boundaries define who decides what in ΔOS. This separation is enforced by system design, not policy.

The Three Authorities

Human Authority

Humans retain ultimate control. This is non-negotiable.

System Authority

The system (ΔOS core) has authority over operational mechanics:

The system cannot:

• Create or modify policies
• Override LIM judgments
• Make value judgments
• Ignore escalation requirements

LIM Authority

LIMs (Ledger-Integrated Modules) have authority over policy evaluation:

LIMs cannot:

• Modify their own policies
• Execute actions directly
• Override other LIMs
• Bypass escalation requirements
• Access resources outside their scope

Authority Boundary Diagram

Enforcement Mechanisms

Authority boundaries are enforced at multiple levels:

Architectural Enforcement

• LIMs cannot access policy modification APIs
• System cannot call action execution APIs
• Audit logs are append-only

Runtime Enforcement

// LIM attempting to modify policy → Error
lim.modifyPolicy(); // Throws: AuthorityViolation

// System attempting to skip escalation → Error
system.forceAllow(escalatedIntent); // Throws: EscalationRequired

// Human override → Always succeeds
human.override(intent, 'allow'); // Succeeds

Audit Enforcement

Every authority exercise is logged:

{
  "type": "authority_exercise",
  "authority": "human",
  "action": "override",
  "target": "int_abc123",
  "from": "block",
  "to": "allow",
  "actor": "user_789",
  "timestamp": "2025-01-25T10:30:00Z",
  "reason": "Business exception approved"
}

Escalation Model

When LIMs cannot make a definitive judgment, authority escalates to humans:

Escalation Triggers

Escalation Flow

LIM returns Escalate
       ↓
System routes to human queue
       ↓
Human receives notification
       ↓
Human makes decision
       ↓
Decision recorded and applied

Escalation SLAs

Escalations have configurable SLAs:

await deltaos.escalation.configure({
  routes: [
    { pattern: 'payment.*', team: 'finance', sla: '15m' },
    { pattern: '*.delete', team: 'data-governance', sla: '1h' }
  ],
  defaultSla: '30m',
  onSlaBreached: 'escalate-to-manager'
});

Override Authority

Humans can override any judgment:

Override Types

Override Recording

All overrides are permanently recorded:

const override = await deltaos.override({
  intentId: 'int_abc123',
  newJudgment: 'allow',
  reason: 'Business exception for Q4 close',
  approval: 'CFO verbal approval, ticket FIN-1234'
});

// Creates immutable audit record
console.log(override.auditRecord);
// {
//   overrideId: 'ovr_xyz789',
//   originalJudgment: 'block',
//   newJudgment: 'allow',
//   reason: 'Business exception for Q4 close',
//   evidence: 'CFO verbal approval, ticket FIN-1234',
//   actor: 'user_456',
//   timestamp: '2025-01-25T10:30:00Z',
//   immutableHash: 'sha256:...'
// }

Kill Switch

The kill switch is always available and cannot be disabled:

Activation

// Any authorized human can activate
await deltaos.killSwitch.activate({
  reason: 'Suspected compromise',
  scope: 'all'  // or specific agents/actions
});

Effect

• All pending Intents → Block
• All new Intents → Block
• Notification to all configured channels
• Audit record created

Recovery

Only humans can deactivate:

await deltaos.killSwitch.deactivate({
  reason: 'Incident resolved, root cause identified',
  approvedBy: 'security-team'
});

Capability Matrix by Authority

See Also

• Autonomy Governance — The governance paradigm
• LIMs — Policy evaluation modules
• Guarantees — System commitments
• Incident Posture — When things go wrong