Skip to content

Security: youngbryan97/aura

SECURITY.md

Security Policy — Aura Cognitive Runtime

Supported Versions

Version Supported
1.0.x ✅ Active
< 1.0 ❌ Research only, no security patches

Reporting a Vulnerability

Do not open a public issue for security vulnerabilities.

Email: security@aura-project.dev PGP Key: See security/pgp-public-key.asc

Response SLA:

  • Acknowledgement: 48 hours
  • Triage: 5 business days
  • Fix (critical): 7 calendar days
  • Fix (high): 14 calendar days
  • Fix (medium/low): next release cycle

Security Architecture

Aura is a locally-deployed AI cognitive runtime. Its threat model differs from cloud SaaS — the primary trust boundary is the local machine, but the AI agent surfaces (tool use, memory, model routing, autonomous action) create unique attack vectors.

Trust Boundaries

┌─────────────────────────────────────────────────┐
│  Local Machine (Operator Trust Boundary)        │
│  ┌───────────────────────────────────────────┐  │
│  │  Aura Runtime Process                     │  │
│  │  ┌─────────────┐  ┌──────────────────┐   │  │
│  │  │ User Input  │  │ Tool/Skill Exec  │   │  │
│  │  │ (untrusted) │  │ (sandboxed)      │   │  │
│  │  └──────┬──────┘  └────────┬─────────┘   │  │
│  │         │                  │              │  │
│  │  ┌──────▼──────────────────▼──────────┐  │  │
│  │  │  Unified Will / AuthorityGateway   │  │  │
│  │  │  (all consequential actions gated) │  │  │
│  │  └──────┬─────────────────────────────┘  │  │
│  │         │                                │  │
│  │  ┌──────▼──────┐  ┌──────────────────┐  │  │
│  │  │ Memory/State│  │ Model Inference  │  │  │
│  │  │ (encrypted) │  │ (local/cloud)    │  │  │
│  │  └─────────────┘  └──────────────────┘  │  │
│  └───────────────────────────────────────────┘  │
│                                                 │
│  ┌───────────────┐  ┌───────────────────────┐  │
│  │ Filesystem    │  │ Network (optional)    │  │
│  │ (workspace)   │  │ (cloud fallback/API)  │  │
│  └───────────────┘  └───────────────────────┘  │
└─────────────────────────────────────────────────┘

Security Controls Summary

Control Status Implementation
Action governance (Will/Authority) ✅ Enforced core/will.py, core/governance/will_gate.py
Tool sandboxing ✅ Enforced security/code_sandbox.py, security/sandbox.py
Input sanitization ✅ Enforced security/sanitizer.py
Memory encryption at rest ✅ Available core/state/vault.py
Secret scanning (CI) ✅ CI gate .github/workflows/enterprise-gate.yml
Dependency vulnerability scanning ✅ CI gate pip-audit in CI
SBOM generation ✅ Available tools/build_provenance.py
Governance bypass detection ✅ Enforced tools/lint_governance.py
Prompt injection defenses ✅ Multi-layer Input sanitizer + integrity checks
Cloud fallback privacy controls ✅ Configurable AURA_CLOUD_FALLBACK_POLICY

Secure Defaults

Aura ships with these defaults in production mode (AURA_MODE=production):

  • All tool/skill execution is sandboxed
  • Self-modification is disabled
  • Cloud fallback requires explicit opt-in
  • Memory writes require Will receipts
  • Unsigned/unmanifested skills do not load
  • Broad filesystem access is denied by default
  • Network access requires explicit skill permission
  • Debug/research endpoints are disabled
  • Verbose logging is disabled (no secret leakage)

Dependency Management

  • Production dependencies are pinned with hashes in requirements/core.txt
  • Lock file: requirements_lock.txt
  • Automated vulnerability scanning via pip-audit and OSV
  • SBOM generated per release via make provenance
  • No fallback dependency installation in production mode

Incident Response

See docs/runbooks/ for operational incident response procedures.

Compliance Mappings

Framework Document
NIST SSDF security/NIST_SSDF_MAPPING.md
OWASP ASVS security/OWASP_ASVS_MAPPING.md
OWASP LLM Top 10 security/OWASP_LLM_MAPPING.md
SLSA security/SLSA_PROVENANCE.md
MITRE ATLAS security/MITRE_ATLAS_MAPPING.md

There aren't any published security advisories