FOR UNDERWRITERS & RISK TEAMS

Insurer Alignment

Runtime enforcement evidence underwriters need to assess autonomous AI risk. Deterministic authorization, denial, and lockdown — with audit-grade proof.

What HALMAI Controls

Deterministic Authorization Gate

Every agent action passes through policy evaluation. No side effects without explicit ALLOW decision.

Exactly-Once Execution

Idempotency guarantees prevent duplicate actions. Each approved action executes once and only once.

Tamper-Evident Audit Trail

SHA-256 hash-chained ledger. Any modification breaks the chain and triggers alerts.

Replay Verification

Re-evaluate past decisions under historical policy snapshots. Produces PASS/FAIL/UNKNOWN results.

Lockdown Mode

Immediate execution halt on critical anomalies. Manual unlock required to resume operations.

Incident Events

Critical events (budget breach, PII leak blocked, loop detected) logged with severity classification.

Underwriter Evidence Pack

A downloadable package containing enforcement artifacts for risk assessment and policy evaluation.

Governance Audit Report (Sample)

Score breakdown, denial rates, anomaly frequency

Decision/Execution Ledger Sample (Redacted)

Hash-chained records demonstrating audit integrity

Policy Rules Snapshot Proof

Versioned rule exports with cryptographic hashes

Incident Log Examples

Blocked actions, lockdown events, severity classification

Control Mapping Document

Controls mapped to evidence artifacts

Download Evidence Pack

Sample package • PDF + JSON

Controls → Evidence Mapping

ControlWhat It PreventsEvidence Produced
Budget Caps / Rate LimitsRunaway spend, recursive loopsBlocked actions + shadow log entries
Egress FiltersSecret/PII leakageDeny reason + audit event
Veto WindowHigh-risk unreviewed actionsHeld decision + veto record
Hash ChainLedger tamperingReplay FAIL + integrity alert
Lockdown ModeContinued execution during incidentLockdown event + manual unlock log
Replay VerificationPolicy drift, inconsistent decisionsPASS/FAIL/UNKNOWN results per decision

Kernel Enforcement Guarantees

No side-effects without ALLOW decision

All agent actions blocked at the kernel level until policy evaluation returns ALLOW.

Exactly-once execution

Idempotency keys and double-execution guards prevent duplicate action execution.

Tenant isolation

Multi-tenant architecture with data and policy separation. One tenant's lockdown does not affect others.

Durable audit events

All decisions and executions persisted to hash-chained ledger before action completion.

Replay verification produces PASS / FAIL / UNKNOWN

Historical decisions can be re-evaluated against versioned policy snapshots. UNKNOWN indicates missing policy version.

UNDERWRITER PACKAGE

Request Underwriter Brief

Receive a custom technical brief tailored for your risk assessment process.

Manual follow-up within 2 business days. No automated scheduling.

Schedule an Underwriter Walkthrough

Technical deep-dive into enforcement evidence for risk teams, underwriters, and compliance officers.

Schedule Walkthrough