AIP-03: Policy Model

Abstract#

Defines risk-tier evaluation and enforcement requirements.

Risk Tiers#

• LOW: auto-executable

• MEDIUM: user confirmation required

• HIGH: explicit approval required; may require multi-signer / MFA for critical categories

Normative Requirements#

1. Every tool call MUST be evaluated against an active policy.

2. HIGH-risk actions MUST require explicit confirmation.

3. Infrastructure-destructive actions SHOULD require multi-factor confirmation or privileged approval.

4. Policy decisions MUST be recorded in audit events.

5. Policy configuration MUST be versioned and hashable (policy_hash).

6. Policy evaluation MUST be applied before side effects.

Compliance Consideration#

Policy enforcement establishes clear authorization boundaries and reduces operational liability.