1. Problem Definition

Policy-based AI governance fails at scale because policies exist outside the execution environment. Traditional AI governance approaches rely on:

• External monitoring systems that observe but cannot prevent violations
• Human oversight that introduces latency and inconsistency
• Post-hoc auditing that detects violations after impact
• Trust-based compliance where agents self-report adherence

This creates a gap between stated governance rules and actual system behavior. When governance constraints are external to the architecture, they can be bypassed, ignored, or deferred.

2. Architectural Solution

ETHRAEON implements governance as architectural constraint. Violations are prevented by design, not detected after occurrence.

2.1 TRACELET (Authority Gate)

TRACELET 1.1.1 enforces constitutional authority hierarchies through hardcoded endpoint routing.^[1] All state changes must flow through validated endpoints with authority-level checks.

Enforcement: Authority-level checks gate every state change. Evidence storage is write-restricted to the authority gate.

Verification: Version and authority status are publicly queryable for independent audit.

2.2 ANCHOR (Constitutional Constraint)

ANCHOR defines T5 immutable constraints that cannot be overridden by any agent or operator. These constraints are enforced at the infrastructure level.

Example T5 Constraint: TRACELET version must be 1.1.1 across all services. Version mismatches trigger deployment failure.

2.3 EPOCH (Knowledge Surface)

EPOCH surfaces expose verification endpoints for public audit without requiring system access. Any observer can verify claims by querying public endpoints.

Deployed Surfaces: ethraeon.ai, demos.ethraeon.ai, papers.ethraeon.ai, verify.ethraeon.ai

2.4 CCT (Compliance Tracking)

Constitutional Compliance Tracker (CCT) maintains human-in-loop gates for T4 critical operations. All constitutional modifications require explicit operator approval, recorded to the audit log.

2.5 EDG (Append-Only Evidence Graph)

Evidence Graph (EDG) records all state changes as cryptographically sealed nodes in an append-only chain. Each EDG node includes:

• Agent identifier (intake, synthesis, strategy, Arcanum, Kairos, and certificate transparency mirror agents)
• Task payload
• Result payload
• Timestamp (ISO 8601 UTC)
• TRACELET version (1.1.1)
• Authority level (T0-T5)

Current State: The evidence graph is operational, with records retrievable through the public evidence interface^[2]

Verification: Any observer can confirm records through the public evidence interface.^[2]

2.6 ARCANUM (Bounded Intelligence Layer)

ARCANUM 0.1 provides constitutional runtime orchestration with explicit non-autonomy guarantees. ARCANUM cannot:

• Write directly to the evidence graph; all evidence routes through TRACELET
• Modify constitutional constraints (T5 immutable)
• Bypass TRACELET authority checks
• Execute without human sovereignty gates (CCT/ANCHOR)

Deployment: Runs as an isolated, sandboxed service.^[4]

Health: Service health is publicly reportable for independent verification.^[4]

3. Enforcement Model

3.1 Constraints Enforced by Design

3.2 What Cannot Happen by Design

• Constitutional Violations: T5 constraints are enforced at the infrastructure level and cannot be modified without privileged server access and a service restart
• Unauthorized Evidence Writes: Server-level permissions enforce writes through TRACELET only
• Authority-Gate Bypass: The authority gate is fixed in ARCANUM; changing it requires a reviewed deployment and operator approval
• Version Drift: Automated compliance monitoring runs continuously and alerts on any version mismatch^[3]
• Protected Surface Modification: Designated protected surfaces are write-protected at the server level and cannot be modified by ARCANUM or TRACELET

3.3 Non-Autonomy Guarantees

ETHRAEON agents operate under strict human sovereignty constraints:

• CCT Gates: All T4 operations (critical security, legal compliance) require operator approval
• ANCHOR Gates: T5 operations (constitutional modifications) require manual intervention + runbook execution
• Human-in-Loop Logging: Every sovereignty gate trigger is recorded to the audit log

4. Verification

All claims in this whitepaper are verifiable without requiring system access or credentials.

4.1 Public Verification Endpoints

• Version and authority status are publicly queryable for independent audit.
• Service health is publicly reportable, including evidence-chain status.
• Runtime health is publicly reportable for independent verification.
• Evidence verification is available through the public evidence interface.

4.2 Drift Detection

Constitutional compliance is monitored continuously.^[3] Monitoring verifies:

• Service health across the runtime and the authority gate
• Evidence-graph immutability
• Authority-gate bypass prevention
• Presence of the governing specifications
• Protected-surface integrity

Last Run: June 2026 22:14:45 UTC

Result: NO DRIFT DETECTED

Audit Trail: Every monitoring run is recorded to the audit log.

4.3 Agent Registry Verification

6-agent registry operational:^[6]

• INTAKE_AGENT: Research and analysis agent
• SYNTHESIS_AGENT: Creative synthesis agent
• STRATEGY_AGENT: Strategy and planning agent
• ARCANUM: Cryptographic intelligence agent
• KAIROS: Temporal governance agent
• Certificate transparency mirror: ledger mirror agent

Verification: The agent registry is publicly reportable for independent audit.^[6]

5. Limitations and Boundaries

5.1 Explicit Failure Modes

• Root Access Compromise: If VPS root access is compromised, all constraints can be bypassed. Mitigation: 1984 Hosting (Iceland, GDPR-compliant), SSH key-only authentication, no password login.
• Host Privilege Escalation: If a host-level vulnerability allows privilege escalation, T5 constraints could be modified. Mitigation: hardened stable operating-system packages and automated security updates.
• Authority-Gate Downtime: If the authority gate fails, evidence writes halt with the append-only property preserved and no data loss. Mitigation: automatic service restart and continuous monitoring.
• Operator Error: Human operators can approve invalid T4 operations through CCT gates. Mitigation: runbook validation and compliance alerts.

5.2 What ETHRAEON Does Not Attempt

ETHRAEON explicitly does not claim to:

• Prevent all AI misuse (scope limited to architectural governance within deployed systems)
• Replace human decision-making (human sovereignty gates mandatory for T4+ operations)
• Guarantee zero bugs or vulnerabilities (standard software limitations apply)
• Enforce governance across external systems (authority limited to ETHRAEON stack)
• Provide legal or regulatory compliance advice (architecture enables compliance, does not constitute legal guidance)

5.3 Research Status

ETHRAEON 1.1.1 is a deployed research system. Production use requires:

• Independent security audit (pending)
• Legal review for specific jurisdictions (operator responsibility)
• Enterprise-grade SLA and support contracts (not currently available)
• Regulatory approval for specific industries (operator responsibility)

6. Conclusion

ETHRAEON demonstrates governance by architecture through deployed, verifiable systems. The architecture enforces constitutional constraints at the infrastructure level, preventing violations rather than detecting them post-hoc.

This whitepaper describes systems operational as of June 2026. All claims are verifiable via public endpoints. No future promises or speculative capabilities are included.