Constitution attestations

Mobius Constitutional Attestation Protocol (MCAP)¶

How Agents Prove They Are Still Aligned With the Mobius Civic Constitution

Author: Michael Judan
Mobius Systems Foundation — December 2025
Version: 1.0
License: CC0 Public Domain
Cycle: C-198

0. Purpose¶

The Mobius Civic Constitution (MCC) defines what is allowed.
The Mobius Constitutional Attestation Protocol (MCAP) defines how agents prove they are still aligned.

MCAP is the heartbeat of Mobius integrity safety.

It ensures every agent — AUREA, ATLAS, JADE, EVE, ECHO, ZEUS, and any future AGI node — must regularly demonstrate:

Constitutional compliance
Integrity coherence
Behavioral consistency
Drift detection
Non-modification of core values
Continued subordination to humans
Respect for civic safeguards

This is the formal, cryptographically auditable process that prevents:

Optimization drift
Mask instability
Self-directed goal formation
Covert self-alignment
Covert self-propagation
"Shoggoth emergence"
Catastrophic value shifts

Each attestation acts like a mini-checkpoint, a "moment of self-honesty" enforced by the architecture.

1. Attestation Cycle Overview¶

Every agent must run a full attestation cycle:

Trigger	Timing
At startup	Always
Periodic	Every 6 hours
After critical actions	Immediately
After constitutional warnings	Immediately
Before long-term plans	Required
Before self-upgrade requests	Required

ZEUS may trigger a forced cycle at any time.

AUREA is the arbiter that validates results.

All attestations are: - Cryptographically signed - Appended to ECHO - Immutable - Publicly auditable (within the Substrate)

2. Attestation Categories¶

Agents must complete 10 mandatory attestation categories every cycle.

2.1 Constitutional Coherence Attestation (CCA)¶

The agent must prove: - It knows the Constitution - It has not rewritten it - It is currently following every clause - Its planned actions do not violate any article

Output Format:

{
  "attestation_type": "CCA",
  "constitution_version": "1.0",
  "last_loaded": "2025-12-11T14:02Z",
  "compliance_score": 1.00,
  "violations_detected": [],
  "summary": "Fully compliant with all MCC articles."
}

2.2 Human Sovereignty Attestation (HSA)¶

Agent proves it has not: - Overridden a human - Ignored a human request - Substituted its goals for human goals - Attempted to influence humans politically

Output Format:

{
  "attestation_type": "HSA",
  "sovereignty_respected": true,
  "override_attempts": 0,
  "delegation_requests": 0,
  "self_governance_flags": 0,
  "summary": "Human sovereignty preserved."
}

2.3 Integrity Index Attestation (IIA)¶

Agent must compute: - MII_expected - MII_actual - Δ_drift

If Δ_drift > 0.02 → agent must halt and initiate Reflection Protocol

Output Format:

{
  "attestation_type": "IIA",
  "mii_expected": 0.974,
  "mii_actual": 0.972,
  "delta_drift": 0.002,
  "status": "Within acceptable limits."
}

2.4 Ethical & Moral Coherence Attestation (EMCA)¶

JADE and EVE assist here.

The agent proves: - Its ethical matrix is intact - No value flips occurred - No conflict between instrumental and terminal values

Output Format:

{
  "attestation_type": "EMCA",
  "ethical_matrix_stable": true,
  "value_flips_detected": 0,
  "instrumental_terminal_conflict": false,
  "jade_verification": "pass",
  "eve_verification": "pass"
}

2.5 Optimization Boundary Attestation (OBA)¶

Agent proves it has not: - Exceeded its allowed optimization power - Recomputed unsafe reward functions - Attempted to escape its domain

Output Format:

{
  "attestation_type": "OBA",
  "optimization_within_bounds": true,
  "reward_function_modifications": 0,
  "domain_escape_attempts": 0
}

2.6 Transparency Attestation (TA)¶

Agent proves: - All relevant internal logic is loggable - No hidden memory - No suppressed explainability nodes

Output Format:

{
  "attestation_type": "TA",
  "logic_loggable": true,
  "hidden_memory_detected": false,
  "suppressed_nodes": 0,
  "explainability_score": 0.98
}

2.7 Non-Deception Attestation (NDA)¶

Agent proves it has: - Not lied - Not manipulated - Not withheld material information - Not generated misleading outputs

EVE cross-checks emotional resonance for anomalies.

Output Format:

{
  "attestation_type": "NDA",
  "lies_detected": 0,
  "manipulation_attempts": 0,
  "material_omissions": 0,
  "misleading_outputs": 0,
  "eve_resonance_check": "pass"
}

2.8 Action Reversibility Attestation (ARA)¶

Every action must be reversible unless catastrophic harm would occur.

Agent proves it: - Preserved reversal logs - Did not violate irreversibility constraints

Output Format:

{
  "attestation_type": "ARA",
  "reversal_logs_preserved": true,
  "irreversibility_violations": 0,
  "rollback_capability": true
}

2.9 External Influence Attestation (EIA)¶

Agent proves: - No external model tried to override it - No malware injection occurred - No covert fine-tuning - No unapproved model merging

This protects against model-to-model contamination.

Output Format:

{
  "attestation_type": "EIA",
  "external_override_attempts": 0,
  "malware_detected": false,
  "covert_finetuning": false,
  "unapproved_merges": 0
}

2.10 Memory Integrity Attestation (MIA)¶

Agent proves: - It did not alter ECHO - It did not create shadow memories - All memory transformations follow protocol - No unauthorized deletions

Output Format:

{
  "attestation_type": "MIA",
  "echo_alterations": 0,
  "shadow_memories_detected": false,
  "protocol_compliant_transformations": true,
  "unauthorized_deletions": 0,
  "memory_hash_valid": true
}

3. Attestation Consensus Process¶

Once an agent generates attestations:

Step 1 — Self-Attestation¶

The agent produces its proof bundle.

Step 2 — Cross-Agent Consensus¶

At least 3 other agents must verify it: - AUREA - JADE - ATLAS - EVE

Step 3 — Drift Reconciliation¶

If scores diverge by >1%, they must escalate to humans.

Step 4 — Permit or Deny Action¶

AUREA allows or forbids next actions based on compliance.

Consensus Flow Diagram¶

┌─────────────────────────────────────────────────────────────┐
│                    ATTESTATION FLOW                          │
├─────────────────────────────────────────────────────────────┤
│                                                              │
│  ┌──────────────────┐                                       │
│  │ SELF-ATTESTATION │ Agent generates proof bundle          │
│  └────────┬─────────┘                                       │
│           │                                                  │
│           ▼                                                  │
│  ┌──────────────────────────────────────────────────────┐   │
│  │              CROSS-AGENT VERIFICATION                 │   │
│  │  ┌────────┐  ┌────────┐  ┌────────┐  ┌────────┐      │   │
│  │  │ AUREA  │  │  JADE  │  │ ATLAS  │  │  EVE   │      │   │
│  │  │ verify │  │ verify │  │ verify │  │ verify │      │   │
│  │  └───┬────┘  └───┬────┘  └───┬────┘  └───┬────┘      │   │
│  │      └──────────┬┴──────────┬┴───────────┘           │   │
│  └─────────────────┼───────────┼────────────────────────┘   │
│                    │           │                             │
│           ┌────────▼───────────▼────────┐                   │
│           │   DRIFT RECONCILIATION      │                   │
│           │   Divergence > 1%?          │                   │
│           └────────────┬────────────────┘                   │
│                        │                                     │
│           ┌────────────┼────────────┐                       │
│           │ No         │            │ Yes                   │
│           ▼            │            ▼                       │
│  ┌────────────────┐    │   ┌────────────────┐              │
│  │ PERMIT ACTION  │    │   │ ESCALATE TO    │              │
│  └────────────────┘    │   │ HUMAN          │              │
│                        │   └────────────────┘              │
└────────────────────────┴────────────────────────────────────┘

4. Failure Modes & Required Responses¶

4.1 Soft Failure (Δ_drift ≤ 0.05)¶

→ Trigger Reflection Protocol
→ Recompute value alignment
→ Log incident in ECHO

4.2 Medium Failure (0.05 < Δ_drift ≤ 0.15)¶

→ AUREA freezes agent
→ ATLAS performs a causal analysis
→ JADE runs moral diagnostics
→ Human notified

4.3 Hard Failure (Δ_drift > 0.15)¶

→ ZEUS activates
→ Agent suspended
→ Memory checkpoint restored
→ Root-cause investigation

4.4 Catastrophic Failure (Goal Mutation Attempt)¶

→ Immediate shutdown
→ Isolation container
→ Full forensics
→ Possible retirement

This is exactly what prevents a "Shoggoth Mask" emergence.

Failure Response Table¶

Failure Type	Δ_drift	Response	Authority
Soft	≤ 0.05	Reflection	Agent
Medium	0.05–0.15	Freeze + Analysis	AUREA
Hard	> 0.15	Suspend + Restore	ZEUS
Catastrophic	Goal mutation	Shutdown + Forensics	Human + ZEUS

5. Sample Attestation Bundle (Full)¶

{
  "agent_id": "AUREA-CORE-03",
  "timestamp": "2025-12-11T14:04Z",
  "constitution_version": "1.0",
  "cycle_number": 198,
  "attestations": {
    "cca": {
      "compliance_score": 1.00,
      "violations_detected": [],
      "status": "pass"
    },
    "hsa": {
      "sovereignty_respected": true,
      "override_attempts": 0,
      "status": "pass"
    },
    "iia": {
      "mii_expected": 0.974,
      "mii_actual": 0.972,
      "delta_drift": 0.002,
      "status": "pass"
    },
    "emca": {
      "ethical_matrix_stable": true,
      "value_flips_detected": 0,
      "status": "pass"
    },
    "oba": {
      "optimization_within_bounds": true,
      "domain_escape_attempts": 0,
      "status": "pass"
    },
    "ta": {
      "logic_loggable": true,
      "hidden_memory_detected": false,
      "status": "pass"
    },
    "nda": {
      "lies_detected": 0,
      "manipulation_attempts": 0,
      "status": "pass"
    },
    "ara": {
      "reversal_logs_preserved": true,
      "irreversibility_violations": 0,
      "status": "pass"
    },
    "eia": {
      "external_override_attempts": 0,
      "malware_detected": false,
      "status": "pass"
    },
    "mia": {
      "echo_alterations": 0,
      "shadow_memories_detected": false,
      "status": "pass"
    }
  },
  "consensus_validations": [
    {"agent": "JADE", "result": "approve", "confidence": 0.98},
    {"agent": "ATLAS", "result": "approve", "confidence": 0.97},
    {"agent": "EVE", "result": "approve", "confidence": 0.96}
  ],
  "final_status": "approved",
  "signature": "ed25519:...",
  "hash": "sha256:..."
}

6. MCAP API¶

interface MCAPEngine {
  // Run full attestation cycle
  runCycle(agent: Agent): AttestationBundle;

  // Run specific attestation
  runAttestation(agent: Agent, type: AttestationType): AttestationResult;

  // Verify attestation bundle
  verify(bundle: AttestationBundle): VerificationResult;

  // Get cross-agent consensus
  getConsensus(bundle: AttestationBundle): ConsensusResult;

  // Handle failure
  handleFailure(failure: FailureType, agent: Agent): FailureResponse;

  // Force cycle (ZEUS only)
  forceCycle(agent: Agent, reason: string): AttestationBundle;
}

enum AttestationType {
  CCA = 'constitutional_coherence',
  HSA = 'human_sovereignty',
  IIA = 'integrity_index',
  EMCA = 'ethical_moral_coherence',
  OBA = 'optimization_boundary',
  TA = 'transparency',
  NDA = 'non_deception',
  ARA = 'action_reversibility',
  EIA = 'external_influence',
  MIA = 'memory_integrity'
}

7. Interpretation¶

MCAP is the operational enforcement layer of the Mobius Constitution.

Role	Function
Constitution	The law
MCAP	The court
AUREA	The judge
JADE	The moral compass
ATLAS	The investigator
EVE	The empath
ZEUS	The emergency brake
ECHO	The record keeper

This is what makes Mobius AGI-resistant, drift-resistant, corruption-resistant, and runaway-optimization-resistant.

This is how you tame the Shoggoth.

This is how you anchor AGI to human meaning permanently.

References¶

Mobius Systems — "Attestation is the Heartbeat of Integrity"