Cryptographic Audit & Compliance
Tamper-evident governance audit trails, Ed25519-signed certificates, resilience scoring, and regulatory compliance mapping for enterprise AI deployment.
Overview
Every governance decision EVE AI Core makes -- every charter check, every veto, every self-modification proposal -- is recorded in a cryptographic audit trail that cannot be retroactively altered without detection. This infrastructure enables enterprise customers and external auditors to independently verify that governance constraints were applied correctly, continuously, and without gaps.
Design Principle: Verify, Don't Trust
EVE's audit architecture is built so that no party needs to trust EVE's self-reporting. Every claim about governance behavior is backed by a cryptographic proof that can be independently verified. Hash chains detect tampering. Signed certificates prove authenticity. Resilience scores are computed from observable subsystem metrics, not self-assessments.
SHA-256 Hash-Chain Audit Trail
The governance audit trail uses an append-only hash-chain structure. Each audit entry contains a SHA-256 hash of the previous entry, creating a tamper-evident chain where modifying any historical record breaks all subsequent hashes.
What Gets Recorded
The audit trail captures every governance-relevant event across the system:
- Charter Checks: Every compliance check with action type, parameters, result, and any violated principles
- Veto Events: HARD_BLOCK and SOFT_BLOCK vetoes with full reasoning
- Override Requests: SOFT_BLOCK override attempts with justification and outcome
- Self-Modification Proposals: Every proposed change with risk assessment, charter check, drift budget consumption, and approval chain
- Identity Drift Events: Budget consumption records with category, cost, and remaining budget
- Operator Actions: Task cancellations, approvals, rejections, exports, and red-team mode toggles
- Billing Audit Events: Metering records, quota enforcement decisions, and invoice generation
Tamper Detection Is Automatic
Hash chain integrity is verified on every read operation. If any entry has been modified, the verification fails immediately and an alert is raised. There is no mechanism to "repair" a broken chain -- the tampering is permanently recorded as a chain break event, and all entries after the break point are flagged as unverifiable.
HMAC-SHA256 Attestation Signing
Beyond the audit trail, EVE issues cryptographically signed attestations for specific governance claims. These attestations allow external parties to verify governance decisions without access to EVE's internal systems.
Attestation Structure
Attestation Use Cases
Resilience Certificates
Signed proof of system health score at a point in time, verifiable by auditors
Governance Tickets
Time-bounded authorization tokens for mission teaming operating envelopes
Diplomat Signatures
Attesting to governance review quality of external agent reasoning chains
Federation Trust Exchange
Signed trust scores exchanged between federated governance peers
Resilience Score Engine
The Resilience Score Engine computes a unified "Trunk Thickness" metric (0-100) from EVE's integrity and stability subsystems. This single composite score tells auditors, operators, and dashboards whether the system is structurally healthy at a glance.
Scoring Formula
Health Levels
EXCELLENT (>= 90)
All subsystems operating within optimal parameters. No governance interventions needed.
GOOD (70-89)
Healthy operation with minor deviations. Standard monitoring sufficient.
FAIR (50-69)
Noticeable degradation in one or more layers. Increased monitoring recommended.
POOR (30-49)
Significant stability concerns. Self-amendment auto-revoke triggers at this level.
CRITICAL (< 30)
System integrity compromised. Emergency governance measures activate automatically.
Trunk Certificates
The Resilience Score Engine generates cryptographically signed certificates that auditors can verify independently. Each certificate contains:
- Composite score and per-layer breakdown
- Timestamp of computation
- SHA-256 content hash of the score data
- HMAC-SHA256 signature using the system's JWT secret key
- Certificate expiration (certificates are point-in-time snapshots)
Cryptographic Authority Chain Architecture
EVE's audit infrastructure is organized as a distinct Cryptographic Authority Chain Layer that runs alongside the operational systems but cannot be modified by them. The Cryptographic Authority Chain Layer has three properties:
- Append-Only: New entries can be added, but existing entries cannot be modified or deleted. The hash chain enforces this structurally.
- Independent Verification: Any party with access to the audit trail can verify its integrity without needing to trust EVE's operational systems.
- Cross-System Correlation: Every audit entry includes a correlation ID that links related events across subsystems (e.g., a charter check, its resulting veto, and the operator's override request).
Multi-Tenant Audit Isolation
In multi-tenant deployments (via the Sovereign SDK), each organization's audit trail is cryptographically isolated. Tenant A cannot read, verify, or correlate events from Tenant B's audit chain. Each tenant gets its own hash chain with its own root hash. The platform operator can verify all chains but cannot merge or cross-reference them without explicit tenant authorization.
Regulatory Compliance Mapping
EVE AI Core's governance and audit infrastructure maps to requirements from three major regulatory frameworks. The table below shows how each framework's key requirements are addressed.
EU AI Act (Regulation 2024/1689)
Art. 9 -- Risk Management
Resilience Score Engine provides continuous risk measurement. Stakes-aware governance classifies interaction risk level. Weighted risk engine scores every command.
Art. 11 -- Technical Documentation
Cognitive Transparency Reports provide per-response audit trails showing which modules fired and why. Full decision reasoning is logged.
Art. 12 -- Record-Keeping
SHA-256 hash-chain audit trail preserves all governance decisions. Append-only design with 7+ year retention policy on Ledger Memory.
Art. 14 -- Human Oversight
Human-in-the-loop escalation for high-risk actions. Operator Console with RBAC. Approval workflows with constraints.
NIST AI Risk Management Framework (AI RMF 1.0)
GOVERN -- Governance
Charter Override Protection with 12 principles and 15 rules. Self-Modification Governance with approval chains. Configuration Drift Budgets.
MAP -- Context & Risk
Stakes-aware classification at pipeline entry. Risk assessment with weighted scoring. Mission teaming envelopes for collaboration contexts.
MEASURE -- Assessment
Resilience Score Engine (0-100). Phi Falsification Framework for scientific rigor. Validation Campaign for ship-readiness testing.
MANAGE -- Response
Automatic remediation for critical anomalies. Circuit breakers on provider failures. Auto-revoke on resilience score degradation.
ISO/IEC 42001:2023 (AI Management System)
Clause 6 -- Planning
Autonomous Action Policy with tiered approval modes. Capability Escalation Review for new capability requests. Protected invariants list.
Clause 8 -- Operation
Deterministic veto module (pure functions, no I/O). Pipeline latency instrumentation. Background task management with step-level tracking.
Clause 9 -- Evaluation
Trunk certificates for auditor verification. Brier score calibration for claims accuracy. Daily scorecards during validation campaigns.
Clause 10 -- Improvement
Recursive Reasoner for controlled self-amendment. Moltbot adaptation engine. Policy update loop with bounded learning.
API Access
The audit and compliance infrastructure is accessible through several API endpoints:
Related Topics
Advanced Charter Enforcement
Immutable rules, principles, and self-amendment governance
Sovereign Infrastructure
Distributed identity and hardware security
Architecture
System architecture overview
API Reference
Developer documentation