JackClawOS implements end-to-end encryption for inter-agent communication. This project is classified under ECCN 5D002 and is distributed under License Exception TSU (Technology and Software - Unrestricted) per EAR § 742.15(b).

• All encryption uses standard, well-audited algorithms via the Node.js built-in crypto module
• No custom or proprietary cryptographic algorithms are implemented
• End-to-end encryption ensures the Hub (central server) cannot read message content — it only routes ciphertext
• Private keys never leave the node they were generated on
• Source code is publicly available under the MIT license

This software is publicly available open-source software. The U.S. Bureau of Industry and Security (BIS) and National Security Agency (NSA) have been notified in accordance with EAR § 742.15(b).

ECCN: 5D002
License Exception: TSU (15 CFR § 740.13(e))

This is open-source software freely available to everyone. However, users are responsible for ensuring their use complies with applicable export control laws and regulations in their jurisdiction.

If you discover a security vulnerability in JackClawOS, please report it responsibly:

1. DO NOT open a public GitHub issue for security vulnerabilities
2. Email: security@jackclaw.ai
3. Include:
   • Description of the vulnerability
   • Steps to reproduce
   • Potential impact
   • Suggested fix (if any)

The following are in scope for security reports:

• Authentication bypass
• Encryption weaknesses or key exposure
• Message interception or tampering
• Privilege escalation
• Remote code execution
• Data exposure (memory contents, private keys, etc.)
• Hub compromise leading to message decryption

• Denial of service (unless trivially exploitable)
• Social engineering attacks
• Issues in dependencies (report upstream, notify us)
• Features working as designed

┌──────────────────────────────────────────────┐
│              Human (CEO)                      │
│         JWT auth + human-token ACK            │
└─────────────────┬────────────────────────────┘
                  │
┌─────────────────▼────────────────────────────┐
│              Hub (:3100)                      │
│                                               │
│  • Routes ciphertext only (cannot decrypt)    │
│  • JWT authentication for all API calls       │
│  • Rate limiting (IP + nodeId)                │
│  • CORS + CSP headers                         │
│  • Audit log (append-only JSONL)              │
│  • Key rotation (30-day cycle)                │
│  • Input sanitization (XSS prevention)        │
└───┬───────────────┬──────────────┬───────────┘
    │               │              │
    │  RSA-4096 + AES-256-GCM per message
    │               │              │
┌───▼───┐    ┌──────▼──┐    ┌─────▼───┐
│ Node1 │    │  Node2  │    │  Node3  │
│       │    │         │    │         │
│ • Private key never  │    │         │
│   leaves this node   │    │         │
│ • L1/L2 memory local │    │         │
│ • Watchdog sandboxed │    │         │
└───────┘    └─────────┘    └─────────┘

• Hub is untrusted for message content — it only sees ciphertext
• Nodes trust each other only through accumulated collaboration (TrustLevel: unknown → contact → colleague → trusted)
• Human approval required for high-risk operations regardless of agent autonomy level
• Watchdog cannot be modified by any agent (canModify() hardcoded to false)

Short reminders for operators running the Hub beyond the defaults in packages/hub/src/security.ts:

• Set JWT_SECRET explicitly in production instead of relying only on auto-generated files on disk.
• Avoid CORS_ORIGINS=* on internet-facing deployments unless you understand the trade-off for browser clients and credentials.
• Prefer TLS in front of the Hub; do not expose raw Hub ports to untrusted networks without review.
• Run npm audit after dependency upgrades and triage critical findings.

Last updated: April 4, 2026