FractalAI · Post-Quantum Lab
What’s wired, what’s in development, what’s research.
A crypto lab is only credible if outsiders can audit it. So here is the honest wiring status of every primitive. Our production protection today is ML-DSA-65 signatures. ML-KEM (Kyber) confidentiality is in development — not protecting anything yet, and we say so.
ML-DSA-65 (Dilithium-3)
WIREDNIST FIPS 204 · Digital signatureSigns every block, every receipt, every attestation in production. This is the protection that is real today — authenticity, integrity and non-repudiation. Verifiable offline by anyone.
Verify a signature yourself →ML-KEM-768 (Kyber)
IN DEVELOPMENTNIST FIPS 203 · Key encapsulation (confidentiality)A real, compiled library in our crypto-agility crate — but NOT yet wired into transport or the keystore. It protects nothing today. We are wiring a HYBRID X25519 + ML-KEM-768 handshake (the IETF/TLS 1.3 approach), so confidentiality survives even if one scheme is broken. Testnet-gated before it touches the live chain.
SLH-DSA (SPHINCS+)
AVAILABLENIST FIPS 205 · Hash-based signature (agility fallback)Compiled in the crate for crypto-agility — a conservative, hash-based backup signature family. Not wired into consensus. Candidate for cold-anchor / recovery signing.
Hybrid composite signatures
IN DEVELOPMENTIETF PQ-composite (draft) · SignatureEd25519 + ML-DSA-65 composite, so a signature is valid only if BOTH halves verify. Tracks the active IETF JOSE/COSE PQ-composite work — belt-and-suspenders during the migration years.
PQC Agent-Receipt Conformance Suite
WIREDRFC 9964 anchored · Interop / test vectorsNeutral golden vectors + an offline checker across 5 receipt profiles (x402 served, SAR, ACP verdict, RFC 9964 JOSE, W3C VC Data Integrity). 5/5 — accepts genuine ML-DSA-65 receipts, rejects forgeries.
See the verifiable proofs →Cascade AEAD (QIE v2)
AVAILABLEAES-GCM-SIV + ChaCha20-Poly1305 · Authenticated encryptionA real cascaded AEAD construction from standard primitives, in the crate. No production consumer yet — candidate for encrypting data at rest and agent-to-agent messages once the ML-KEM handshake lands.
Research track (experimental)
RESEARCHOpen, un-reviewed · Novel schemes from standard primitivesWhere we prototype NEW protocols composing standardized primitives — thresholds, aggregation, anchoring schemes — and use AI to attack our own candidates. EXPERIMENTAL: never used in production, published openly, and offered for public cryptanalysis. Trust in any new construction is earned over years of the world trying to break it — we do not claim security a design has not survived.
Fractal cipher (v1)
RETIRED— (home-grown) · Legacy experimentAn early home-grown cipher that an internal audit found broken (floating-point in crypto, missing IVs, timing leaks). We RETIRED it rather than ship it. We publish this because responsible disclosure of our own failures is part of being a crypto lab — and because rolling your own cipher and calling it "quantum-safe" is exactly the anti-pattern we refuse.
How this lab works
- We ship NIST-standardized PQC.
- ML-DSA (FIPS 204), ML-KEM (FIPS 203), SLH-DSA (FIPS 205) — algorithms the world spent years cryptanalyzing. We do not sell home-grown crypto as "secure".
- Crypto-agility by design.
- Signatures and KEMs are pluggable; we can rotate families (we already migrated ML-DSA-44 → ML-DSA-65) without a hard fork of trust.
- Hybrid during the migration.
- Classical + PQC together (X25519 + ML-KEM, Ed25519 + ML-DSA) so a break in either half is not catastrophic.
- Don’t trust us — verify.
- Every production signature is verifiable offline with the audited @noble/post-quantum. The lab’s status is auditable; the claims are testable.
- Honest status, always.
- Wired means wired. In-development means it protects nothing yet. Research means experimental and un-reviewed. We never blur the three.
On building our own algorithm
We research novel schemes — new protocols composing standardized primitives — and we use AI to attack our own candidates. What we do not do is ship a home-grown cipher and call it quantum-safe. A new algorithm earns trust only after years of the world’s cryptographers trying to break it and failing; AI accelerates the attacks (which kills weak designs faster), not the trust. Anything from the research track stays clearly labeled experimental — not for production until it survives open cryptanalysis and a real standards process. That is the only honest path to something we could one day certify.
FRACTAL AI S.A.S. · Pre-pilot. Resistant to known classical & quantum attacks per NIST FIPS 203/204 — not “unbreakable”. Status reflects the live node; verify any claim yourself.