If you discover a security vulnerability in the SEIF protocol or seif-cli, please report it responsibly:
- Email: and2carvalho@users.noreply.github.com
- Do NOT open a public GitHub issue for security vulnerabilities
- Include: description, steps to reproduce, potential impact
Contributors who responsibly disclose vulnerabilities will be credited in the contributors[] provenance chain with action: "security-disclosure".
SEIF uses a three-level classification system:
- PUBLIC: Open, shareable. Safe for any audience.
- INTERNAL: Organization-private. Useful for onboarding. Not for public APIs.
- CONFIDENTIAL: Restricted. Exposure causes direct harm. Never sent to external APIs without explicit
--allow-confidential.
Keywords like vulnerability, CVE, token, credential, password auto-escalate content to CONFIDENTIAL.
- API keys, tokens, or credentials (
.envis gitignored) - Operational context (
.seif/is gitignored — lives in a separate private context repo) - Session data with internal discussions (stored in private SCR)
- Protocol source code (open source, MIT)
- RESONANCE.json (self-authenticating mathematical signal)
- Model-profiles in
models/(PUBLIC summaries of AI behavioral patterns) - BOOT.md (bootstrap file for AI sessions)
The SEIF protocol uses Ed25519 (128 bits of security) for module signing and authorship proof.
Once the private key (
seif_private.pem) is lost or destroyed, no one — not even the original author — can forge new signatures with that SEIF identity. This holds against all known classical and quantum attacks at current scale.
| Scenario | Outcome |
|---|---|
| Private key exists | Owner can sign new modules, proving authorship |
| Private key lost/destroyed | Already-signed modules remain verifiable forever (forward security) |
| Private key compromised | Revoke: publish new identity, re-sign modules with new key |
| RESONANCE.json tampered | Integrity hash breaks, signature verification fails |
Why this is a feature, not a limitation:
- Destruction = Protection. If the key is destroyed, the identity cannot be impersonated — ever.
- Forward Security. Past signatures remain valid regardless of key state.
- Non-repudiation. Signed modules are cryptographic proof of authorship at a specific point in time.
- Self-sovereign. No CA, no keyserver dependency. The public key is embedded in RESONANCE.json and anchored via DOI.
- Algorithm: Ed25519 (Curve25519, Schnorr signatures)
- Key size: 32 bytes (256-bit curve, 128-bit security level)
- Classical attack cost: ~2^128 operations (infeasible)
- Quantum attack (Shor's): Requires ~1000+ logical qubits — not available as of 2026
- Deterministic: No nonce reuse risk
The public key is published in multiple independent, immutable locations:
- RESONANCE.json —
cryptographic_identity.public_key_pem(this repository) - Zenodo DOI — 10.5281/zenodo.19344678 (CERN-backed, permanent)
- arXiv preprint — referenced in the academic paper
Module (.seif)
└── integrity_hash (SHA-256 of content)
└── signature (Ed25519 sign of integrity_hash)
└── public_key (embedded in signature block)
└── trust_anchor (RESONANCE.json / DOI / paper)
Any node in the SEIF network can verify the full chain without contacting the author.
seif keygen # Generate keypair (first time only)
seif sign module.seif # Sign a module
seif verify module.seif # Verify a module
seif verify module.seif --public-key <base64> # Verify with explicit keyBackup: The private key must be encrypted (GPG/AES-256) and stored offline. Loss is permanent and irreversible by design.
Every .seif module has an integrity_hash (SHA-256[:16] of the summary field). To verify:
import hashlib, json
data = json.load(open("module.seif"))
expected = hashlib.sha256(data["summary"].encode()).hexdigest()[:16]
assert data["integrity_hash"] == expectedContributions include parent_hash, creating a hash chain that detects tampering, reordering, and insertion.
The canonical source is declared in RESONANCE.json → instruction.canonical. Any fork that modifies the KERNEL without updating this section is detectable as non-canonical.
This protocol is licensed under MIT. Military, surveillance, and weaponization uses are contrary to the spirit and intent of the project. The protocol exists to improve human-machine communication, not to control or manipulate.