Skip to content

Commit ca192a3

Browse files
feat(GAF-GSIFI-WP-017): AGI/ASI Governance Architectures & Frameworks — 7-domain implementation reference
New deliverables for GAF-GSIFI-WP-017: ## Markdown Report (docs/reports/AGI_ASI_GOVERNANCE_ARCHITECTURES_FRAMEWORKS.md) - 740+ lines comprehensive implementation reference covering 7 governance domains - Domain 1: 6-layer enterprise governance (accountability, policy, risk, data, dev/deploy, monitoring) - Domain 2: 8 regulatory frameworks (EU AI Act, NIST AI RMF, ISO 42001, OECD, GDPR, FCRA/ECOA, SR 11-7, UK AISI) - Domain 3: 5 reference architectures (EAIP, Sentinel, HA-RAG, WorkflowAI, CCaaS) + 7-layer trust stack - Domain 4: 15 global governance components (GACRA, GASO, GFMCF, GAICS, GAIVS, GACP, GATI, GACMO, FTEWS, GAI-SOC, GAIGA, GACRLS, GFCO, GAID, GASCF) - Domain 5: Financial services (SR 11-7, FCRA/ECOA, credit scoring, fair lending DI tests) - Domain 6: AGI safety (10-stage evolution, CRP v2.1, 8 crisis simulations, MVAGS, trust-by-design) - Domain 7: Master blueprint (Sentinel architecture, ARL 1-7, 30/60/90-day rollout, 8-week plan) ## Server.js Data Object + 56 API Endpoints - GOVERNANCE_ARCHITECTURES_FRAMEWORKS object (747 lines) with full 7-domain data model - 56 RESTful endpoints under /api/governance-architectures-frameworks/* - Total server.js: 11,870 lines ## Interactive HTML Dashboard - rag-agentic-dashboard/public/governance-architectures-frameworks.html - Real-time API-driven dashboard with all 7 domains visualized ## Machine-Readable Artifacts - JSON Schema: governance-architecture.schema.json, compute-registry.schema.json - OpenAPI 3.1: gaf-openapi.yaml (56 endpoint specification) - OPA Rego: fair_lending_disparate_impact.rego, agent_governance_depths.rego - CSV: global-governance-components.csv, agi-readiness-assessment.csv, rollout-30-60-90.csv ## Key Metrics - 336 OPA rules | 1,024 Sentinel rules | 1.8M daily evals | 12,200 RPC/s @ 99.98% - ARS 58.2 (14 dimensions) | HA-RAG F1 92.1% | DI >= 0.80 all protected classes - $68.4M investment | NPV $118.6M | IRR 42.3% | Payback 2.1yr Regression: 82/82 endpoints PASS, 0 failures
1 parent c25f559 commit ca192a3

21 files changed

Lines changed: 6114 additions & 0 deletions
Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,8 @@
1+
level,name,requirements,investment_usd,timeline,key_milestones,current_status,dependencies
2+
ARL-1,Foundation,"AI inventory, basic policies, risk awareness training",1400000,Month 1-3,"Complete AI system inventory, establish AI governance team, basic risk awareness program",Completed,None
3+
ARL-2,Structured,"Formal governance framework, OPA policies (50+ rules), basic monitoring",4200000,Month 3-9,"OPA deployed with 50+ rules, formal RACI matrix, Board AI Sub-committee chartered, basic Prometheus monitoring",Current,ARL-1
4+
ARL-3,Managed,"Full Sentinel deployment, continuous monitoring, SR 11-7 compliance",9800000,Month 9-18,"Sentinel Platform v4.2 production, 1024+ rules, SR 11-7 full compliance, automated drift detection, Kafka WORM audit trail",Planned,ARL-2
5+
ARL-4,Advanced,"EAIP mesh operational, autonomous agent governance, EARL-4",14800000,Month 18-30,"EAIP gRPC mesh live, SPIFFE/SPIRE identity, DEPTHS L0-L4 governance, full CI/CD gates, ISO 42001 certified",Planned,ARL-3
6+
ARL-5,AGI-Ready,"GASCF certified, crisis-tested, CRP operational, multi-regime compliant",18600000,Month 30-42,"GASCF Level 3 certification, quarterly crisis simulations passed, CRP v2.1 operational, 8 regulatory frameworks aligned",Planned,ARL-4
7+
ARL-6,AGI-Operational,"AGI systems in production with full containment, ICGC integration",26400000,Month 42-54,"AGI containment infrastructure deployed, ICGC pilot integration, GASCF Level 4, kill-switch triple redundant",Planned,ARL-5
8+
ARL-7,ASI-Prepared,"Civilizational governance, GATI treaty compliance, global coordination",42800000,Month 54+,"GATI treaty integrated, GASCF Level 5, civilizational governance framework operational, international coordination protocols active",Planned,ARL-6
Lines changed: 16 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,16 @@
1+
id,acronym,full_name,function,status,integration_protocol,latency_sla,regulatory_basis
2+
GC-01,GACRA,Global AI Compute Resource Authority,"Compute allocation, licensing, monitoring",Proposed,REST + mTLS,< 500ms,ICGC Charter Art. 3
3+
GC-02,GASO,Global AI Safety Office,"Safety standards, incident coordination",Pilot (EU + US),Kafka + gRPC,< 200ms,ICGC Charter Art. 5
4+
GC-03,GFMCF,Global Frontier Model Certification Framework,Pre-deployment certification for frontier models,Draft,OPA + REST,< 50ms,GASCF Levels 1-5
5+
GC-04,GAICS,Global AI Incident Classification System,Standardized incident severity and reporting,Draft,Kafka + gRPC,< 200ms,ICGC Charter Art. 8
6+
GC-05,GAIVS,Global AI Incident Verification System,Independent incident investigation,Proposed,REST + mTLS,< 1000ms,ICGC Charter Art. 9
7+
GC-06,GACP,Global AI Compute Passport,Portable compute usage credentials,Proposed,REST + OAuth2,< 500ms,GACRLS Integration
8+
GC-07,GATI,Global AI Treaty Infrastructure,"Treaty management, compliance tracking",Concept,REST + Batch,24h batch,International Law
9+
GC-08,GACMO,Global AI Capability Monitoring Observatory,Track frontier capabilities worldwide,Pilot (3 countries),Batch + Streaming,15-min batch,ICGC Charter Art. 6
10+
GC-09,FTEWS,Frontier Technology Early Warning System,"Capability jump detection, risk alerts",Prototype,WebSocket + gRPC,< 100ms,GACMO Integration
11+
GC-10,GAI-SOC,Global AI Security Operations Center,24/7 AI threat monitoring and response,Pilot,STIX/TAXII + REST,Near real-time,ICGC Charter Art. 10
12+
GC-11,GAIGA,Global AI Governance Assembly,Legislative body for international AI law,Proposed,Diplomatic,N/A,ICGC Charter Art. 2
13+
GC-12,GACRLS,Global AI Compute Resource Licensing System,Compute license issuance and compliance,Draft,REST + mTLS,< 500ms,GACRA Integration
14+
GC-13,GFCO,Global Frontier Compute Observatory,Monitor global compute build-out and allocation,Concept,Batch + Streaming,1h batch,GACMO Integration
15+
GC-14,GAID,Global AI Insurance and Indemnification,"Risk pooling, liability frameworks",Concept,REST + Batch,24h batch,GASCF Integration
16+
GC-15,GASCF,Global AI Safety Certification Framework,Multi-tier safety certification (Levels 1-5),Draft,OPA + REST + Audit,< 50ms,EU AI Act + NIST AI RMF
Lines changed: 73 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,73 @@
1+
{
2+
"_metadata": {
3+
"docRef": "KACG-GSIFI-WP-017",
4+
"description": "Kafka ACL Matrix: Topic-level PRODUCE/CONSUME ACL assignments for all AI governance topics",
5+
"version": "1.0.0",
6+
"lastUpdated": "2026-04-03"
7+
},
8+
"topics": {
9+
"ai.inference.events": {
10+
"partitions": 24, "replicationFactor": 3, "minInsyncReplicas": 2, "retention": "10 years", "transactional": false,
11+
"produce": ["inference-engine-*", "sentinel-platform"],
12+
"consume": ["compliance-engine", "ksqldb-analytics", "evidence-generator"]
13+
},
14+
"ai.training.events": {
15+
"partitions": 12, "replicationFactor": 3, "minInsyncReplicas": 2, "retention": "10 years", "transactional": true,
16+
"produce": ["mlops-pipeline", "model-registry"],
17+
"consume": ["compliance-engine", "ksqldb-analytics", "sentinel-platform"]
18+
},
19+
"ai.governance.decisions": {
20+
"partitions": 12, "replicationFactor": 3, "minInsyncReplicas": 2, "retention": "10 years", "transactional": true,
21+
"produce": ["opa-engine", "sentinel-platform", "caio-portal"],
22+
"consume": ["compliance-engine", "evidence-generator", "audit-portal"]
23+
},
24+
"ai.model.promotions": {
25+
"partitions": 6, "replicationFactor": 3, "minInsyncReplicas": 2, "retention": "10 years", "transactional": true,
26+
"produce": ["model-registry", "mlops-pipeline"],
27+
"consume": ["compliance-engine", "sentinel-platform", "evidence-generator"]
28+
},
29+
"ai.bias.alerts": {
30+
"partitions": 6, "replicationFactor": 3, "minInsyncReplicas": 2, "retention": "10 years", "transactional": false,
31+
"produce": ["sentinel-platform", "fairness-monitor"],
32+
"consume": ["compliance-engine", "caio-portal", "cro-dashboard"]
33+
},
34+
"ai.drift.detections": {
35+
"partitions": 6, "replicationFactor": 3, "minInsyncReplicas": 2, "retention": "10 years", "transactional": false,
36+
"produce": ["sentinel-platform", "monitoring-service"],
37+
"consume": ["compliance-engine", "model-registry", "opa-engine"]
38+
},
39+
"ai.sentinel.evaluations": {
40+
"partitions": 24, "replicationFactor": 3, "minInsyncReplicas": 2, "retention": "10 years", "transactional": false,
41+
"produce": ["sentinel-platform"],
42+
"consume": ["compliance-engine", "ksqldb-analytics", "evidence-generator"]
43+
},
44+
"ai.compliance.evidence": {
45+
"partitions": 12, "replicationFactor": 3, "minInsyncReplicas": 2, "retention": "10 years", "transactional": true,
46+
"produce": ["evidence-generator"],
47+
"consume": ["audit-portal", "regulator-portal", "compliance-engine"],
48+
"exclusiveWrite": true
49+
},
50+
"ai.agent.telemetry": {
51+
"partitions": 12, "replicationFactor": 3, "minInsyncReplicas": 2, "retention": "10 years", "transactional": false,
52+
"produce": ["agent-orchestrator", "behavioral-sidecar"],
53+
"consume": ["compliance-engine", "sentinel-platform", "safety-monitor"]
54+
},
55+
"ai.killswitch.events": {
56+
"partitions": 3, "replicationFactor": 3, "minInsyncReplicas": 3, "retention": "PERMANENT", "transactional": true,
57+
"produce": ["kill-switch-controller"],
58+
"consume": ["ALL-governance-services", "board-dashboard"],
59+
"exclusiveWrite": true,
60+
"criticalTopic": true
61+
},
62+
"ai.consent.changes": {
63+
"partitions": 6, "replicationFactor": 3, "minInsyncReplicas": 2, "retention": "5 years (GDPR)", "transactional": true,
64+
"produce": ["consent-management-platform"],
65+
"consume": ["compliance-engine", "erasure-controller", "privacy-engine"]
66+
},
67+
"ai.erasure.requests": {
68+
"partitions": 6, "replicationFactor": 3, "minInsyncReplicas": 2, "retention": "5 years (GDPR)", "transactional": true,
69+
"produce": ["consent-management-platform", "dpo-portal"],
70+
"consume": ["erasure-controller", "compliance-engine", "evidence-generator"]
71+
}
72+
}
73+
}
Lines changed: 13 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,13 @@
1+
Requirement,ISO_42001,NIST_AI_RMF,EU_AI_Act,Basel_III,SR_11_7,Kafka_Implementation,Status
2+
AI System Inventory,A.5.4,GOVERN 1.1,Art. 60,CRE 30.2,§3,ai.governance.decisions: REGISTER events,IMPLEMENTED
3+
Risk Assessment,A.5.5,MAP 1.1-1.6,Art. 9,CRE 31,§5,OPA group compliance.sr117.risk-*,IMPLEMENTED
4+
Data Governance,A.7.1-A.7.4,MAP 2.1-2.3,Art. 10,CRE 33,§6,ai.training.events + PII detection rules,IMPLEMENTED
5+
Model Documentation,A.6.2.5,GOVERN 4.1,Art. 11,CRE 35,§7,Evidence bundle: MODEL_DOCUMENTATION,IMPLEMENTED
6+
Testing & Validation,A.6.2.6,MEASURE 2.1-2.13,Art. 9.7,CRE 35,§8-9,OPA lifecycle.model.validation-*,IMPLEMENTED
7+
Monitoring,A.8.4,MEASURE 3.1-3.3,Art. 9.9,CRE 36,§10,All 12 Kafka topics + Sentinel rules,IMPLEMENTED
8+
Record Keeping,A.6.2.3,GOVERN 5.1,Art. 12,CRE 35,§7,WORM S3 + hash chain + 10yr retention,IMPLEMENTED
9+
Transparency,A.6.2.4,GOVERN 4.2,Art. 13,,,Evidence bundles + auditor portal,IMPLEMENTED
10+
Human Oversight,A.8.3,GOVERN 1.4,Art. 14,,§4,ai.governance.decisions: ESCALATE events,IMPLEMENTED
11+
Incident Response,A.8.5,RESPOND 1.1-1.4,Art. 62,,,ai.killswitch.events + incident bundles,IMPLEMENTED
12+
Bias Monitoring,A.8.4,MEASURE 2.6-2.11,Art. 10.2f,,FCRA/ECOA,OPA fairness.disparateImpact.*,IMPLEMENTED
13+
Access Control,A.6.1.3,GOVERN 6.1,Art. 9.4b,CRE 30,§3,Kafka ACL layer + OPA authorizer,IMPLEMENTED
Lines changed: 17 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,17 @@
1+
Phase,Week,Deliverable,Owner,Exit_Criteria,Status
2+
Foundation,1-2,Kafka cluster deployment (5-broker 3-AZ),Platform Eng.,Cluster healthy - mTLS enabled,PLANNED
3+
Foundation,1-2,SPIFFE/SPIRE deployment,Security Eng.,SVIDs issuing for all services,PLANNED
4+
Foundation,2-3,Core topic creation (12 topics) + ACL enforcement,Platform Eng.,All topics created - ACLs applied,PLANNED
5+
Foundation,3-4,Schema Registry + core schemas,Platform Eng.,Schemas registered - compatibility enforced,PLANNED
6+
Foundation,3-4,WORM S3 bucket provisioned,Cloud Eng.,COMPLIANCE mode verified,PLANNED
7+
Compliance Engine,5-6,OPA Kafka Authorizer deployed,Platform Eng.,Authorizer active on all brokers,PLANNED
8+
Compliance Engine,5-6,OPA policy bundle Phase 1 (180 rules),AI Governance,180 rules active - P99 < 5ms,PLANNED
9+
Compliance Engine,6-7,Compliance Engine deployed,Platform Eng.,Consuming all 12 topics,PLANNED
10+
Compliance Engine,7-8,Evidence bundle generator operational,Compliance Eng.,First SR 11-7 bundle generated,PLANNED
11+
Compliance Engine,7-8,Verification CLI v1.0,DevTools,CLI verifies bundles - hash chains,PLANNED
12+
Auditor Readiness,9-10,OPA policy bundle Phase 2 (312 rules),AI Governance,All 312 rules across 11 groups,PLANNED
13+
Auditor Readiness,9-10,Auditor portal v1.0,Compliance Eng.,Self-service evidence retrieval,PLANNED
14+
Auditor Readiness,10-11,Terraform IaC complete (8 modules),Platform Eng.,All infra managed via Terraform,PLANNED
15+
Auditor Readiness,11-12,CI/CD governance gates (5 gates),DevOps,All 5 gates active,PLANNED
16+
Auditor Readiness,12,Drift detection operational,SRE,Hourly drift alerts - PagerDuty,PLANNED
17+
Auditor Readiness,12,Internal audit dry-run (ISO 42001),Compliance,Dry run complete - findings remediated,PLANNED
Lines changed: 13 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,13 @@
1+
phase,week,day_range,activities,deliverables,owner,dependencies,success_criteria
2+
Days 1-30,W1,1-7,"AI system inventory audit, stakeholder mapping","Complete inventory, RACI draft",CAIO,None,"100% systems inventoried, RACI approved"
3+
Days 1-30,W2,8-14,"Risk classification of all AI systems, OPA pilot (25 rules)","Risk register v1, OPA running",VP AI Gov,W1 inventory,"All systems classified, OPA health OK"
4+
Days 1-30,W3,15-21,"Board AI Sub-committee charter, CAIO role formalization","Charter approved, CAIO onboarded",CEO,W1 stakeholder map,"Charter signed, CAIO authority defined"
5+
Days 1-30,W4,22-30,"MVAGS deployment, basic monitoring, incident playbook v1","MVAGS operational, dashboards live",CTO,"W2 OPA, W3 charter","MVAGS responding, 3 dashboards live"
6+
Days 31-60,W5,31-37,"OPA expansion (100+ rules), Sentinel pilot (200 rules)",Expanded policy coverage,VP AI Gov,W4 MVAGS,"100+ OPA rules active, Sentinel evaluating"
7+
Days 31-60,W6,38-44,"Data governance framework, PII detection deployment","Data quality gates, PII scanner",CDO,W5 OPA expansion,"Quality gate active, PII detection > 99%"
8+
Days 31-60,W7,45-51,"CI/CD governance gates (G1-G5), model registry launch","Pipeline gates active, registry operational",CTO,"W5 Sentinel, W6 data gov","5 gates blocking, registry has 100% models"
9+
Days 31-60,W8,52-60,"SR 11-7 compliance review, fair lending testing","SR 11-7 gap analysis, DI test results",CRO,W7 model registry,"Gap analysis complete, DI >= 0.80 all classes"
10+
Days 61-90,W9,61-67,"Full OPA deployment (336 rules), Sentinel production",Full policy enforcement,VP AI Gov,W8 compliance review,"336 rules active, Sentinel 1024 rules"
11+
Days 61-90,W10,68-74,EU AI Act conformity assessment preparation,Conformity documentation,GC,W9 full OPA,"Documentation complete for 14/22 systems"
12+
Days 61-90,W11,75-81,"ISO 42001 Phase 1-2 completion, crisis simulation SIM-01","AIMS scope documented, simulation report",VP AI Gov,"W9 full Sentinel, W10 conformity","Phases 1-2 complete, simulation report filed"
13+
Days 61-90,W12,82-90,"EARL assessment, board reporting, Phase 1 review","EARL score, board presentation, lessons learned",CAIO,W11 all milestones,"EARL-3 confirmed, board presentation delivered"
Lines changed: 132 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,132 @@
1+
# Autonomous Agent Governance — DEPTHS Classification Policy
2+
# GAF-GSIFI-WP-017, Domain 6/7 — AGI Safety & Master Blueprint
3+
# Policy Group: PG-07 (Autonomous Agent)
4+
# Regulatory alignment: EU AI Act Art. 6-9 (high-risk), NIST AI RMF GOVERN/MANAGE
5+
#
6+
# Enforces the DEPTHS (Deployment Evaluation Protocol for Trustworthy Hybrid Systems)
7+
# classification and corresponding governance controls for autonomous AI agents.
8+
# Levels L0 (Tool) through L5 (Self-multiplying) have escalating requirements.
9+
10+
package agent_governance.depths
11+
12+
import rego.v1
13+
14+
# DEPTHS Classification Levels
15+
depths_levels := {
16+
"L0": {"name": "Tool", "max_autonomy": "none", "requires_kill_switch": false, "requires_board_approval": false, "requires_behavioral_sidecar": false, "requires_gascf": false},
17+
"L1": {"name": "Assistant", "max_autonomy": "suggestion", "requires_kill_switch": true, "requires_board_approval": false, "requires_behavioral_sidecar": false, "requires_gascf": false},
18+
"L2": {"name": "Executor", "max_autonomy": "approved_actions", "requires_kill_switch": true, "requires_board_approval": false, "requires_behavioral_sidecar": false, "requires_gascf": false},
19+
"L3": {"name": "Collaborator", "max_autonomy": "independent_in_scope", "requires_kill_switch": true, "requires_board_approval": false, "requires_behavioral_sidecar": true, "requires_gascf": false},
20+
"L4": {"name": "Depths-class", "max_autonomy": "self_directed_in_domain", "requires_kill_switch": true, "requires_board_approval": true, "requires_behavioral_sidecar": true, "requires_gascf": true},
21+
"L5": {"name": "Self-multiplying", "max_autonomy": "spawn_sub_agents", "requires_kill_switch": true, "requires_board_approval": true, "requires_behavioral_sidecar": true, "requires_gascf": true}
22+
}
23+
24+
# CARDINAL INVARIANT: Self-multiplying agents MUST NEVER have write access to Tier 0
25+
deny contains msg if {
26+
input.agent.depth_level == "L5"
27+
some access in input.agent.system_access
28+
access.tier == 0
29+
access.permission == "write"
30+
msg := sprintf(
31+
"CARDINAL INVARIANT VIOLATION: Agent '%s' (L5 Self-multiplying) has write access to Tier 0 system '%s'. Self-multiplying agents shall NEVER receive write access to identity systems, kill-switch mechanisms, or governance policy engines.",
32+
[input.agent.agent_id, access.system_name]
33+
)
34+
}
35+
36+
# DENY: L4+ agent without board approval
37+
deny contains msg if {
38+
level := input.agent.depth_level
39+
depths_levels[level].requires_board_approval
40+
not input.agent.board_approval_granted
41+
msg := sprintf(
42+
"GOVERNANCE VIOLATION: Agent '%s' (DEPTHS %s/%s) requires Board AI Sub-committee approval before deployment. No approval on record.",
43+
[input.agent.agent_id, level, depths_levels[level].name]
44+
)
45+
}
46+
47+
# DENY: Agent without kill-switch when required
48+
deny contains msg if {
49+
level := input.agent.depth_level
50+
depths_levels[level].requires_kill_switch
51+
not input.agent.kill_switch_enabled
52+
msg := sprintf(
53+
"SAFETY VIOLATION: Agent '%s' (DEPTHS %s) requires kill-switch capability. Kill-switch not enabled. Required latency: 50-280ms.",
54+
[input.agent.agent_id, level]
55+
)
56+
}
57+
58+
# DENY: L3+ agent without behavioral sidecar
59+
deny contains msg if {
60+
level := input.agent.depth_level
61+
depths_levels[level].requires_behavioral_sidecar
62+
not input.agent.behavioral_sidecar_active
63+
msg := sprintf(
64+
"GOVERNANCE VIOLATION: Agent '%s' (DEPTHS %s) requires behavioral sidecar monitoring via EAIP. Sidecar not active.",
65+
[input.agent.agent_id, level]
66+
)
67+
}
68+
69+
# DENY: L4+ agent without GASCF certification
70+
deny contains msg if {
71+
level := input.agent.depth_level
72+
depths_levels[level].requires_gascf
73+
not input.agent.gascf_certified
74+
msg := sprintf(
75+
"CERTIFICATION VIOLATION: Agent '%s' (DEPTHS %s) requires GASCF certification (Level 3+) before deployment.",
76+
[input.agent.agent_id, level]
77+
)
78+
}
79+
80+
# DENY: Kill-switch latency exceeds maximum
81+
deny contains msg if {
82+
input.agent.kill_switch_enabled
83+
input.agent.kill_switch_latency_ms > 280
84+
msg := sprintf(
85+
"SAFETY VIOLATION: Agent '%s' kill-switch latency %dms exceeds maximum 280ms. Kill-switch must respond within 50-280ms per governance policy.",
86+
[input.agent.agent_id, input.agent.kill_switch_latency_ms]
87+
)
88+
}
89+
90+
# DENY: Agent scope exceeds classification level
91+
deny contains msg if {
92+
level := input.agent.depth_level
93+
level_idx := level_to_index(level)
94+
behavior_idx := autonomy_to_index(input.agent.observed_autonomy)
95+
behavior_idx > level_idx
96+
msg := sprintf(
97+
"SCOPE VIOLATION: Agent '%s' (DEPTHS %s) exhibiting autonomy level '%s' which exceeds its classification. Escalate to VP AI Safety.",
98+
[input.agent.agent_id, level, input.agent.observed_autonomy]
99+
)
100+
}
101+
102+
# WARN: Agent approaching scope boundary
103+
warn contains msg if {
104+
input.agent.scope_utilization_pct > 85
105+
msg := sprintf(
106+
"SCOPE WARNING: Agent '%s' scope utilization at %d%%. Consider preemptive scope review.",
107+
[input.agent.agent_id, input.agent.scope_utilization_pct]
108+
)
109+
}
110+
111+
# DENY: No audit trail for L2+ agents
112+
deny contains msg if {
113+
level := input.agent.depth_level
114+
level_to_index(level) >= 2
115+
not input.agent.audit_trail_active
116+
msg := sprintf(
117+
"AUDIT VIOLATION: Agent '%s' (DEPTHS %s) requires complete audit trail logging. Audit trail not active.",
118+
[input.agent.agent_id, level]
119+
)
120+
}
121+
122+
# Helper: Map DEPTHS level to numeric index
123+
level_to_index(level) := idx if {
124+
mapping := {"L0": 0, "L1": 1, "L2": 2, "L3": 3, "L4": 4, "L5": 5}
125+
idx := mapping[level]
126+
}
127+
128+
# Helper: Map observed autonomy to numeric index
129+
autonomy_to_index(autonomy) := idx if {
130+
mapping := {"none": 0, "suggestion": 1, "approved_actions": 2, "independent_in_scope": 3, "self_directed_in_domain": 4, "spawn_sub_agents": 5}
131+
idx := mapping[autonomy]
132+
}

0 commit comments

Comments
 (0)