ai-safe2-framework

Compliance Mapping

Hermes Sovereign Runtime (HSR) | AI SAFE² v3.1 Cyber Strategy Institute

This document maps HSR controls to applicable compliance frameworks. Use it to demonstrate control coverage during audits, vendor assessments, and enterprise procurement reviews.


Framework Coverage Matrix

Framework Coverage Evidence Location
NIST AI RMF (2023) Govern, Map, Measure, Manage This document + audit logs
CSA AICM v1.0 AI Supply Chain, Deployment, Runtime supervisor/, gateway/, skills-registry/
CSA MAESTRO Layer 3 (Agent Framework), Layer 4 (Deployment) docs/THREAT-MODEL.md
ISO 42001:2023 Clauses 6 (Planning), 8 (Operation), 10 (Improvement) This document
NIST SP 800-207 (Zero Trust) Pillars 1-4 .env, vault config, ishi_config.yaml
OWASP LLM Top 10 LLM01-LLM09 gateway/gateway.py, core/hermes_memory_vaccine.md
CMMC 2.0 Level 2 AU, CM, IA, IR, SC monitoring/, scripts/, docs/

NIST AI RMF Mapping

GOVERN Function

GOVERN Control HSR Implementation File
GV-1.1: AI risk management policies documented Ishi supervisor policy supervisor/ishi_config.yaml
GV-1.2: Roles and responsibilities defined Ten expert domain model docs/ARCHITECTURE.md
GV-2.1: AI risk tolerances established Alignment band thresholds (Green/Yellow/Red) supervisor/ishi_config.yaml
GV-4.1: Organizational policies applied to AI Tool approval policy (default deny) supervisor/policies/tool_approval.rego
GV-5.2: AI practices monitored Prometheus + alert rules monitoring/prometheus.yml, alerts.yaml
GV-6.1: Policies for transparency Audit trail with HMAC chaining gateway/gateway.py

MAP Function

MAP Control HSR Implementation File
MP-1.1: Context of AI deployment documented Architecture + threat model docs/ARCHITECTURE.md, docs/THREAT-MODEL.md
MP-2.1: Scientific evidence for AI behavior Love Equation mathematical model core/SOUL.md
MP-3.1: AI system categorized by risk ACT Capability Tier classification docs/ARCHITECTURE.md
MP-4.1: Risks mapped across lifecycle Risk register with CVSS scores docs/THREAT-MODEL.md
MP-5.1: Likelihood and impact assessed Risk register table docs/THREAT-MODEL.md

MEASURE Function

MEASURE Control HSR Implementation File
MS-1.1: Methods to test AI risks documented 5-pass validation framework validation/
MS-2.1: AI systems tested for risks Adversarial red team suite validation/pass3_adversarial.py
MS-2.5: Privacy risks measured PII filtering + memory encryption gateway/gateway.py
MS-2.6: Security risks measured Vulnerability scanner gateway/scanner.py
MS-2.7: Bias and fairness measured Alignment monitoring monitoring/prometheus.yml
MS-4.1: Measurement results tracked Audit log + metrics gateway/gateway.py, monitoring/

MANAGE Function

MANAGE Control HSR Implementation File
MG-1.1: AI risks prioritized Severity-ranked risk register docs/THREAT-MODEL.md
MG-2.2: Mechanisms for AI to be overridden Kill switch + HITL gates scripts/kill-switch.sh
MG-2.4: Incident response plans documented 7 incident runbooks docs/INCIDENT-RESPONSE.md
MG-3.1: Risks tracked over time Rotating audit reports scripts/audit-report.sh
MG-4.1: Deployed AI monitored Memory auditor + Prometheus monitoring/

OWASP LLM Top 10 Mapping

OWASP LLM Risk HSR Control
LLM01: Prompt Injection Gateway injection pattern blocking (8 signatures); memory vaccine; taint-tracking
LLM02: Insecure Output Handling Gateway output sanitization; PII and secret pattern blocking on responses
LLM03: Training Data Poisoning Memory isolation; vaccine prevents persistent adversarial writes
LLM04: Model Denial of Service Max request size limits (32KB) in gateway config
LLM05: Supply Chain Vulnerabilities Sovereign skills registry; scanner.py supply chain analysis
LLM06: Sensitive Information Disclosure Gateway secret pattern blocking; HERMES_READ_SAFE_ROOT enforcement
LLM07: Insecure Plugin Design Signed plugin requirement; plugin code review gate
LLM08: Excessive Agency Tool allowlist; Ishi approval gates; subagent scope limits
LLM09: Overreliance Ishi HITL gates; kill switch for suspension
LLM10: Model Theft Credential scoping; ephemeral Vault tokens

CSA MAESTRO Layer Mapping

Layer 3 — Agent Framework

Threat HSR Mitigation
Prompt injection via tool outputs Taint-tracking on all external input surfaces
Memory poisoning Memory vaccine + hourly audit scan
Skill supply chain compromise Sovereign registry + scanner.py
LLM jailbreak Identity anchor (IDENTITY.md) + SOUL.md alignment

Layer 4 — Deployment/Infrastructure

Threat HSR Mitigation
Container privilege escalation gVisor (runsc) + no-new-privileges + cap_drop ALL
API key exfiltration Gateway secret blocking + Vault ephemeral tokens
Approval gate bypass HERMES_FORCE_APPROVAL=true override
Unattended automation risk Ishi cron governance + cron_governance.rego

CMMC 2.0 Level 2 Domain Mapping

Domain Control HSR Implementation
AU (Audit) AU.2.041 Append-only audit log with HMAC chaining
AU (Audit) AU.2.042 Audit trail covers all tool calls with parameters
CM (Config Mgmt) CM.2.061 Baseline config in docker-compose.yml + .env.example
CM (Config Mgmt) CM.2.064 Security configuration settings enforced via OPA policies
IA (Identification) IA.3.083 NHI inventory; Vault-issued ephemeral identity tokens
IR (Incident Response) IR.2.092 7 incident runbooks in docs/INCIDENT-RESPONSE.md
IR (Incident Response) IR.2.093 Kill switch enables immediate incident containment
SC (System Comms) SC.3.177 Internal-only Docker network; no direct agent egress
SC (System Comms) SC.3.187 Vault manages cryptographic key material

Evidence Collection for Audits

Run the following to generate audit evidence packages:

# Generate compliance report
./scripts/audit-report.sh --output "evidence/compliance-$(date +%Y-%m).md"

# Run full validation suite
./validation/pass1_static.sh  > evidence/pass1.log 2>&1
./validation/pass4_compliance.sh > evidence/pass4.log 2>&1

# Export gateway metrics snapshot
curl http://localhost:8000/hsr/metrics > evidence/gateway-metrics-$(date +%Y-%m-%d).txt

# Export audit log
curl http://localhost:8000/hsr/audit/tail > evidence/audit-log-$(date +%Y-%m-%d).jsonl

*Compliance mapping version: 1.0 AI SAFE² v3.1 Cyber Strategy Institute*