Skip to content

Security: mainnebula/Lightbox-Project

Security

SECURITY.md

Lightbox Security Model

What Lightbox Provides

Lightbox is a flight recorder for AI agents that creates tamper-evident logs of tool executions. It provides:

  1. Append-only logging: Events can only be added, never modified
  2. Hash chain integrity: Each event includes the hash of the previous
  3. Offline verification: No network required to verify a session
  4. Content integrity: Any modification is detectable

What Lightbox Does NOT Provide

  1. Confidentiality: Event logs are stored in plaintext
  2. Authentication: No proof of WHO created the log
  3. Non-repudiation: Without signatures, origin cannot be proven
  4. Secure storage: Files use standard filesystem permissions

Threat Model

Threats Addressed

Threat Mitigation
Accidental corruption Hash chain detects any bit changes
Naive tampering Hash mismatch immediately detected
Event deletion Chain break detected at deletion point
Event insertion Chain break at insertion point
Event reordering prev_hash mismatches reveal reordering
Truncation (crash) Distinct from tampering via incomplete line detection

Threats NOT Addressed (v1)

Threat Status
Malicious operator (full log replacement) Not addressed - requires external anchoring
Selective disclosure attacks Not addressed - no signatures
Timing attacks Not addressed - timestamps are observational
Side-channel attacks Not addressed

Verification

lightbox verify <session>

Exit Codes

Code Meaning
0 Valid - hash chain intact
1 Tampered - content modified or chain broken
2 Truncated - incomplete write detected (likely crash)
3 Parse error - invalid JSON
4 Not found

What Verification Checks

  1. Each event's hash matches its computed hash
  2. Each event's prev_hash matches the previous event's hash
  3. First event has prev_hash=null
  4. No truncated/incomplete lines (except crash recovery)

What Verification Does NOT Check

  1. Timestamps are accurate
  2. Tool outputs are truthful
  3. Events are complete (no dropped events before recording)
  4. Log hasn't been entirely replaced

Future Security Enhancements

Planned (not in v1)

  1. Per-session signatures: Sign the final hash to prove authorship
  2. External anchoring: Publish session hashes to immutable stores
  3. Timestamping: Third-party timestamp services for time proofs

Hash Algorithm

v1 uses SHA-256. This provides:

  • 256-bit collision resistance
  • Widely supported and audited
  • Sufficient for integrity verification

Recommendations

For Operators

  1. Protect the ~/.lightbox directory with appropriate permissions
  2. Back up sessions to immutable storage for long-term evidence
  3. Consider external anchoring for high-stakes applications

For Auditors

  1. Verify sessions immediately upon receipt
  2. Compare session hashes out-of-band when possible
  3. Check for truncation warnings (may indicate interrupted writes)

For Developers

  1. Don't store secrets in tool inputs/outputs - use redaction
  2. Consider size limits for large payloads
  3. Use the content_hashes field when redacting for verifiability

Responsible Disclosure

For security issues, please email security@uselightbox.app rather than opening a public issue.

There aren’t any published security advisories