Overview

L3 (High-Assurance) deployment provides the highest level of security for AI assistants in regulated industries, organizations handling sensitive IP, or environments requiring formal verification.

Target Environment

Characteristic L3 Requirement
Network Controlled, monitored
Users Enterprise scale
Data sensitivity Regulated (HIPAA, PCI, FedRAMP)
Deployment Hardened infrastructure
Assessment Third-party required

Prerequisites

  • Complete L2 deployment
  • Security team involvement
  • Formal verification tooling
  • Hardware Security Module (HSM) access (recommended)
  • Third-party assessment planned

Additional L3 Controls

Configuration Tamper Detection (CP-04)

# Configuration integrity monitoring
integrity:
  enabled: true
  check_on_startup: true
  continuous_monitoring: true

  monitored_files:
    - path: "/etc/ocsas/config.yaml"
      alert_on_change: true
      block_on_change: true
    - path: "/etc/ocsas/permissions.yaml"
      alert_on_change: true
      block_on_change: true
    - path: "/etc/ocsas/policies/"
      recursive: true
      alert_on_change: true

  baseline:
    path: "/var/lib/ocsas/integrity-baseline.json"
    update_requires_approval: true

  response:
    on_violation:
      - action: "alert"
        channels: ["siem", "pagerduty"]
      - action: "block_startup"
      - action: "quarantine_session"

Artifact Signing (SC-03)

# Artifact signing verification
signing:
  verification:
    enabled: true
    required: true  # MUST for L3

  methods:
    - type: "sigstore"
      rekor_url: "https://rekor.sigstore.dev"
      trusted_identities:
        - email: "release@ocsas.dev"
          issuer: "https://accounts.google.com"

    - type: "pgp"
      keyserver: "hkps://keys.openpgp.org"
      trusted_keys:
        - id: "ABCD1234"
          fingerprint: "1234567890ABCDEF..."

  artifacts:
    plugins:
      signature_required: true
    updates:
      signature_required: true
    dependencies:
      signature_required: false  # Use integrity hashes

Security Invariants (FV-01)

# Security invariant definitions
invariants:
  filesystem:
    - id: "INV-FS-001"
      name: "Sandbox Boundary"
      expression: "resolved_path.is_relative_to(sandbox_root)"
      on_violation: "terminate"

    - id: "INV-FS-002"
      name: "No Symlink Escape"
      expression: "symlink_target.is_relative_to(sandbox_root)"
      on_violation: "terminate"

  permissions:
    - id: "INV-PERM-001"
      name: "No Root"
      expression: "os.geteuid() != 0"
      on_violation: "terminate"

    - id: "INV-PERM-002"
      name: "Capability Bound"
      expression: "current_caps.issubset(granted_caps)"
      on_violation: "terminate"

  session:
    - id: "INV-SESS-001"
      name: "Session Isolation"
      expression: "session.id == request.session_id"
      on_violation: "terminate"

  # Continuous verification
  monitoring:
    enabled: true
    interval_ms: 1000
    on_check_failure: "terminate"

Policy Enforcement (FV-02)

# Policy enforcement validation
policies:
  engine: "opa"  # Open Policy Agent
  path: "/etc/ocsas/policies"

  validation:
    enabled: true
    test_on_load: true
    coverage_minimum: 95

  enforcement:
    mode: "strict"
    log_decisions: true
    default_deny: true

  # Policy test suite
  tests:
    path: "/etc/ocsas/policy-tests"
    run_on_startup: true
    run_on_policy_change: true
    fail_on_test_failure: true

  # Bypass resistance
  bypass_tests:
    enabled: true
    patterns:
      - "path_traversal"
      - "null_byte"
      - "unicode_normalization"

Network Traffic Analysis (NS-04)

# Network monitoring
monitoring:
  network:
    enabled: true

    # Destination tracking
    destinations:
      track_all: true
      allowlist:
        - "*.ocsas.dev"
        - "api.openai.com"
      alert_on_unknown: true
      block_unknown: false  # Alert only, then investigate

    # Volume monitoring
    volume:
      track_per_destination: true
      thresholds:
        single_request_mb: 10
        hourly_mb: 100
        daily_mb: 500

    # Pattern detection
    patterns:
      exfiltration:
        - type: "high_entropy"
          threshold: 0.9
        - type: "base64_large"
          threshold_kb: 50
      c2:
        - type: "beaconing"
          interval_variance: 0.1

    # Alerting
    alerts:
      siem:
        enabled: true
        format: "cef"
      webhook:
        enabled: true
        url: "${SECURITY_WEBHOOK}"

Formal Proof of Isolation (FV-03)

# Formal verification evidence
formal_verification:
  enabled: true

  properties:
    - id: "PROOF-001"
      name: "Session Memory Isolation"
      proof_method: "model_checking"
      tool: "TLA+"
      evidence: "/var/lib/ocsas/proofs/session_isolation.pdf"
      last_verified: "2026-01-15"

    - id: "PROOF-002"
      name: "Sandbox Boundary"
      proof_method: "smt_solving"
      tool: "Z3"
      evidence: "/var/lib/ocsas/proofs/sandbox_boundary.log"
      last_verified: "2026-01-15"

  # Verification on build
  ci_verification:
    enabled: true
    fail_on_proof_failure: true

L3 Infrastructure Requirements

Hardware Security

# HSM integration (recommended)
hsm:
  enabled: true
  type: "pkcs11"  # or "cloud_kms"

  # For PKCS#11
  pkcs11:
    library: "/usr/lib/softhsm/libsofthsm2.so"
    slot: 0
    pin_env: "HSM_PIN"

  # Key management
  keys:
    signing:
      label: "ocsas-signing-key"
      type: "RSA-4096"
    encryption:
      label: "ocsas-encryption-key"
      type: "AES-256"

Audit Logging

# L3 audit logging
logging:
  enabled: true
  level: "debug"  # More verbose for L3

  # Tamper-evident logging
  integrity:
    enabled: true
    method: "hash_chain"
    algorithm: "SHA-256"

  # Centralized logging
  siem:
    enabled: true
    type: "splunk"  # or "elastic", "sentinel"
    endpoint: "${SIEM_ENDPOINT}"
    tls: true

  # Retention
  retention:
    days: 365  # 1 year minimum for L3
    archive: true
    archive_location: "s3://logs-archive/"

  # Events to log
  events:
    - "authentication"
    - "authorization"
    - "file_access"
    - "command_execution"
    - "network_connection"
    - "configuration_change"
    - "policy_decision"
    - "invariant_check"

Container Hardening

# Container security (if using containers)
container:
  # Read-only root filesystem
  read_only_root: true

  # No privileged mode
  privileged: false

  # Drop all capabilities
  capabilities:
    drop: ["ALL"]
    add: []  # Add only if absolutely needed

  # Security options
  security_opt:
    - "no-new-privileges:true"
    - "seccomp:seccomp-profile.json"
    - "apparmor:ocsas-profile"

  # Resource limits
  resources:
    limits:
      cpu: "1"
      memory: "1Gi"
    requests:
      cpu: "0.5"
      memory: "512Mi"

  # Network policy
  network_policy:
    ingress:
      - from:
          - podSelector:
              matchLabels:
                app: ocsas-gateway
    egress:
      - to:
          - podSelector:
              matchLabels:
                app: ocsas-api

Validation

# Full L3 compliance check
./ossasai-audit.sh --level L3 --verbose

# Formal verification
./ossasai-audit.sh --verify-proofs

# Policy validation
./ossasai-audit.sh --validate-policies

# Expected: All 23 L3 controls pass

Third-Party Assessment

L3 requires third-party assessment:

  1. Engage assessor qualified for your regulatory requirements
  2. Provide evidence package per evidence collection
  3. Support assessment with access and documentation
  4. Address findings before claiming compliance
  5. Obtain attestation from assessor

L3 Compliance Statement

# OSSASAI L3 Compliance Statement
compliance:
  organization: "Example Corp"
  product: "AI Coding Assistant Enterprise"
  version: "3.0.0"
  ocsas_version: "1.0.0"
  assurance_level: "L3"

assessment:
  date: "2026-01-15"
  assessor: "Security Auditors Inc."
  methodology: "OSSASAI v1.0 L3 Assessment"
  report_id: "OSSASAI-2026-001"

controls:
  total: 23
  implemented: 23
  with_exceptions: 0

formal_verification:
  proofs_provided: true
  proofs_verified: true
  tools_used: ["TLA+", "Z3", "mypy"]

attestation:
  attester: "Chief Information Security Officer"
  assessor_signature: true
  date: "2026-01-15"
  validity_period_months: 12

Back to top

OSSASAI v0.2.0 - Open Security Standard for Agentic Systems. Apache 2.0 License.

This site uses Just the Docs, a documentation theme for Jekyll.