Frontier technology · NIST-standardized PQC on the Aurigraph platform · FIPS 203 ML-KEM, FIPS 204 ML-DSA, FIPS 205 SLH-DSA · hybrid post-quantum TLS live at the edge
The cryptography that secures today's digital ledgers — RSA, ECDSA, Diffie-Hellman — rests on mathematical problems that a sufficiently large quantum computer can solve efficiently. Once such a machine exists, every signature ever produced and every ciphertext ever recorded becomes forgeable or readable. For a distributed ledger, whose entire value proposition is that its history cannot be rewritten, this is an existential rather than incremental risk. Aurigraph addresses it by building its cryptographic core on the NIST-standardized post-quantum algorithms ML-DSA-87 (module-lattice digital signatures, FIPS 204) and ML-KEM-1024 (module-lattice key encapsulation, FIPS 203), implemented in-process against the BouncyCastle 1.86 lightweight API and defaulting to NIST Category 5 — the highest standardized security category. The owner's target is that every public-key operation on the platform uses FIPS 203 (ML-KEM), FIPS 204 (ML-DSA) or FIPS 205 (SLH-DSA), and that every classical algorithm still in use is named as an exception. This paper describes the quantum threat, the NIST standards that answer it, Aurigraph's concrete PQC stack, exactly where post-quantum cryptography is and is not applied on the platform, the optional quantum-entropy path, and the migration and crypto-agility strategy. It is deliberately explicit about what is live in production, what is merged in the V12 source but not yet active, and what is only planned, so the reader can trust the claims that are made. Status labels used throughout: Live in production (verified against dlt.aurigraph.io), Merged in V12 source (code exists; not necessarily enabled or deployed), Rolling out, Planned.
dlt.aurigraph.io
Checked against Aurigraph-DLT V12 source at b0515c472 (the commit running in production on 2026-10-22) and, for the rows marked Live in production, against dlt.aurigraph.io on 2026-10-19 and again on 2026-10-20 (OpenSSL 3.6.4, three runs each). "Merged in V12 source" means the code exists on V12; it does not mean production has it enabled. The owner's directive is that everything moves to FIPS 203–205; the rows below name every place that has not yet arrived.
V12
b0515c472
dlt.aurigraph.io:443
dlt.aurigraph.io:9004
QuantumCryptoServiceImpl
PqcWalletService
src/main/java
REQUIRE_MLDSA
SIMULATE
SecP384r1MLKEM1024
X25519MLKEM768
scripts/consensus-tls-native-probe/run.sh --native
SLHDSASigner
consensus_vote_sig_phase
CrossChainProofBridge
real-zk-not-implemented-fail-closed
QuantumConsensus
src/main
f1114a2c5
CryptoOps
IN_PROCESS
GRPC
aurigraph.crypto.qrng.enabled
QRNG_AGENT_ENABLED
QuantumCryptoServiceStub
Public-key cryptography works because certain problems are easy in one direction and infeasible in reverse for a classical computer. RSA relies on the difficulty of factoring large integers; ECDSA and Diffie-Hellman rely on the discrete-logarithm problem over elliptic curves. Both underpin essentially all blockchain signing and key exchange in production today.
Two quantum algorithms undermine these foundations:
The most important consequence is not in the future but in the present. In a "harvest now, decrypt later" (HNDL) attack, an adversary records encrypted traffic and signed data today and simply waits for a capable quantum computer to arrive. Any secret with a long confidentiality horizon — tokenized property rights, legal contracts, medical records, treasury balances, an immutable ledger of transactions meant to stand for decades — is already exposed to an adversary who is patient. This is why migration timing is driven by data lifetime plus attacker capacity, not by the arrival date of the quantum computer alone: data that must remain secure for twenty years must be protected with post-quantum cryptography twenty years before the machine exists.
For a ledger, the threat is compounded. A forged historical signature does not merely leak data; it can be used to assert that a transaction which never happened did happen, attacking the integrity guarantee that is the entire reason the ledger exists.
In August 2024, after nearly a decade of open competition and public cryptanalysis, the U.S. National Institute of Standards and Technology (NIST) published its first finalized post-quantum standards. Aurigraph builds directly on them:
The lattice schemes (ML-KEM, ML-DSA) base their security on the hardness of Module Learning With Errors (Module-LWE) and related module-lattice problems, for which no efficient classical or quantum algorithm is known. NIST security categories express the target: Category 5 — the level Aurigraph defaults to — corresponds to the difficulty of a brute-force key search against AES-256, the strongest of NIST's five levels.
FIPS 204's ML-DSA-87 provides EUF-CMA security (existential unforgeability under adaptive chosen-message attack). Its trade-off relative to ECDSA is size, not security: ML-DSA public keys and signatures are measured in kilobytes rather than tens of bytes. On a distributed ledger this manifests as larger transaction and block payloads — an engineering cost that is well understood and, for enterprise settlement volumes, entirely manageable.
Aurigraph's real post-quantum cryptography runs in-process inside the platform, built on the BouncyCastle 1.86 lightweight API (bcprov / bcpkix / bcutil-jdk18on:1.86, pinned in aurigraph-v12/pom.xml).
bcprov
bcpkix
bcutil-jdk18on:1.86
aurigraph-v12/pom.xml
Signatures — ML-DSA-87, NIST Level 5 (Merged in V12 source). The signing primitive across the crypto subsystem is ML-DSA at parameter set 5 (ML-DSA-87), using MLDSAKeyPairGenerator, MLDSASigner and MLDSAParameters.ml_dsa_87. QuantumCryptoServiceImpl exposes key generation, signing, and verification, alongside ML-KEM encapsulation and SLH-DSA (FIPS 205) key generation, signing and verification.
MLDSAKeyPairGenerator
MLDSASigner
MLDSAParameters.ml_dsa_87
Key encapsulation — ML-KEM-1024, NIST Level 5 (Merged in V12 source). For establishing shared secrets, QuantumCryptoServiceImpl provides ML-KEM-1024 encapsulate/decapsulate through MLKEMKeyPairGenerator, MLKEMGenerator and MLKEMExtractor.
MLKEMKeyPairGenerator
MLKEMGenerator
MLKEMExtractor
Per-user wallets — PqcWalletService (Merged in V12 source). Each user is provisioned a post-quantum wallet whose keypair is ML-DSA-87 at NIST L5, generated with MLDSAKeyGenerationParameters and MLDSAParameters.ml_dsa_87. Addresses are derived with Keccak-256 (last 20 bytes of the hash of the public key). Wallet persistence is handled separately by the wallet RocksDB service.
MLDSAKeyGenerationParameters
Level-5 by default. Both aurigraph.crypto.dilithium.security-level and aurigraph.crypto.kyber.security-level default to "5", yielding ML-DSA-87 / ML-KEM-1024 out of the box. The property names predate the FIPS naming. Crucially, BouncyCastle derives verification parameters from the key object's own embedded parameters, not from this configuration value — so raising the configured level is a safe forward flip and previously issued keys continue to verify unchanged.
aurigraph.crypto.dilithium.security-level
aurigraph.crypto.kyber.security-level
"5"
Native-image engineering — the JCA bypass. Aurigraph compiles toward a GraalVM native image, where the Java Cryptography Architecture (JCA) provider service map is frozen at build time and cannot register new algorithm providers at runtime. To keep post-quantum algorithms available in that environment, the crypto subsystem deliberately calls the BouncyCastle lightweight API directly (MLDSASigner, MLDSAKeyPairGenerator, the ML-KEM classes) rather than going through Cipher.getInstance(...)/Signature.getInstance(...).
Cipher.getInstance(...)
Signature.getInstance(...)
Standard-name migration (done). The earlier code used BouncyCastle's pre-FIPS CRYSTALS-Dilithium, Kyber and SPHINCS+ classes. The migration to the FIPS-named ML-DSA, ML-KEM and SLH-DSA classes (in org.bouncycastle.crypto.*) is complete: no pqc.legacy class remains in src/main/java, and the hash-based path uses FIPS 205 SLH-DSA (SLHDSASigner, AV11-3552). Some class names (Dilithium5SignatureService) and the two configuration property names keep the old terminology.
org.bouncycastle.crypto.*
pqc.legacy
Dilithium5SignatureService
Performance. Signature and key sizes have been measured: an ML-DSA-87 signature is 4,627 bytes and a public key 2,592 bytes (platform repository docs/perf/PQC-MLDSA87-LOAD-2026-10.md). Latency and throughput are not measured yet: the runs so far were on a contended host, so no figure is published. A quiet-host run and external verification are scheduled (AV11-3955).
docs/perf/PQC-MLDSA87-LOAD-2026-10.md
Real PQC on the platform lives in the public TLS edge, the crypto, wallet and consensus subsystems:
Public edge TLS (Live in production). dlt.aurigraph.io:443 and the gRPC port :9004 terminate TLS 1.3 with hybrid X25519MLKEM768 key exchange (ML-KEM-768 per FIPS 203, combined with X25519). This protects session keys against harvest-now-decrypt-later for clients that offer ML-KEM. Note that ML-KEM-768 is NIST Category 3, not Category 5, and that the handshake is still authenticated with a classical ECDSA P-256 certificate.
:9004
Consensus log entries (Implemented, not active in production). Every replicated Raft log entry is signed and verified with ML-DSA-87 (AV11-3757), with a persistent per-node identity key (AV11-3792) and a ceil(2N/3) quorum. The code is shipped in the production build, but the path is not exercised while production consensus runs in SIMULATE mode on a single standalone node.
PQC wallets (Merged in V12 source). PqcWalletService provisions ML-DSA-87 wallet keypairs. EVM-compatible wallets still use secp256k1 ECDSA keys (a named classical exception, below), so not every wallet is post-quantum.
Credential and key material (Merged in V12 source). Keypair generation, signing, verification, and ML-KEM key establishment go through QuantumCryptoServiceImpl at NIST L5. ML-KEM-1024 is used narrowly: five source files call the BouncyCastle ML-KEM classes directly (the self-test, the two quantum crypto services, the HSM key manager and the benchmark harness).
Consensus votes (Rolling out). VoteSignatureService accepts ML-DSA-87 vote envelopes in DUAL_ACCEPT phase and still accepts Ed25519. The REQUIRE_MLDSA phase is gated on every registered validator holding a real ML-DSA-87 key and has not been flipped in production; since AV11-3543 the flip is persisted and one-way. It is therefore not correct to say every consensus message is ML-DSA-signed.
VoteSignatureService
DUAL_ACCEPT
Ricardian contracts (Merged in V12 source). Contract signing requires an ML-DSA-87 key for the signer DID (AV11-3761). Multi-party signing is being fixed (AV11-3840), and anchoring to the ledger is not yet done (AV11-3863).
The CryptoOps consolidation boundary (Rolling out). A unifying CryptoOps interface (default algorithm ML-DSA-87 / L5) routes some operations through a single seam, with the canonical Battua wallet/crypto service as the intended remote implementation over gRPC.
Removed simulation scaffold. An earlier consensus class named QuantumConsensus was a simulation scaffold that performed no real lattice cryptography. It has been removed from the production source tree (commit f1114a2c5, AV11-3553), and a guard test, NoSimulatedCryptoGuardTest, blocks simulated crypto from returning to src/main. Real post-quantum cryptography is the QuantumCryptoService / PqcWalletService path described in §3 plus the ML-DSA-87 vote path above.
NoSimulatedCryptoGuardTest
QuantumCryptoService
Other limits that remain, stated plainly.
Named classical exceptions (2026-10-20). These are still in use and are not post-quantum: the edge TLS certificate (ECDSA P-256); the classical X25519 / secp256r1 fallback for clients without ML-KEM; Ed25519, still accepted for consensus votes in DUAL_ACCEPT phase and for transaction signatures over gRPC; Ed25519 on KYC records; an ephemeral ECDSA key on DID cross-chain attestations (AV11-3853); secp256k1 ECDSA for EVM-compatible wallets; RS256 and HS256 API tokens. ISO 20022 messages carry no ML-DSA co-signature yet (AV11-3855), and Know-Your-Agent signs with ML-DSA-65 (Category 3), below the Category 5 default.
Aurigraph includes a seam for a Quantum Random Number Generator (QRNG) entropy source, via QuantumEntropyClient and the IBM-QRNG-backed HttpQuantumEntropySource, allowing key material to be seeded from a hardware quantum entropy source. Status: optional, off by default.
QuantumEntropyClient
HttpQuantumEntropySource
This path is disabled by default and fail-safe by design. The QRNG client is governed by aurigraph.crypto.qrng.enabled, which defaults to false, and the separate QRNG agent is governed by QRNG_AGENT_ENABLED, which also defaults to false. When disabled, unreachable, timing out, or returning a short or non-200 response — under any failure whatsoever — the client returns nothing and the platform falls back to a classical cryptographically secure pseudo-random number generator (CSPRNG). The design principle is explicit: a compromised or unavailable QRNG can never weaken key material, only ever supplement it. The IBM upstream is similarly wrapped so callers never block on its availability.
We therefore make no claim of live quantum entropy in the default configuration. The platform's default randomness is a classical CSPRNG; QRNG is an optional, opt-in enhancement whose failure modes are all safe.
Post-quantum migration is not a one-time switch but an ongoing discipline, and Aurigraph is structured for it:
The intended end state consolidates all key custody, signing, and PQC behind the Battua canonical crypto service, with the rest of the platform — and mobile clients — acting as gRPC consumers of that hardened, single home for cryptography. That end state is planned, not shipped.
Roadmap items describe planned and in-progress direction, not shipped guarantees.
The quantum threat to public-key cryptography is not speculative — it is a known consequence of Shor's algorithm, and "harvest now, decrypt later" makes it a present obligation for any system whose data must outlive the arrival of a large quantum computer. A distributed ledger, built to make history permanent, has the least room of all to get this wrong. Aurigraph's V12 source implements the finalized NIST algorithms — ML-DSA-87 signatures (FIPS 204) and ML-KEM-1024 key encapsulation (FIPS 203) at NIST Level 5 — in-process on BouncyCastle 1.86, with a deliberate JCA bypass so the code survives native-image compilation. It is not yet true that every platform message is post-quantum protected: consensus votes accept ML-DSA-87 alongside Ed25519, hybrid ML-KEM TLS is live only at the public edge (with a classical certificate and a classical fallback), consensus-layer ML-DSA-87 signing is implemented but not active in production, ML-KEM is used in only a few places, and SLH-DSA (FIPS 205) is implemented but no production signing path calls it. This paper draws a clear line between what is live, what is rolling out and what is planned, and it declines to claim live quantum entropy where the default is a fail-safe classical CSPRNG. That honesty is not a caveat on the security posture; it is part of it. For the ledger, consensus, and tokenization layer this cryptography protects, see the Aurigraph DLT whitepaper; for the canonical wallet and crypto service it consolidates toward, see the Battua whitepaper.
bcutil-jdk18on
Explore: About · Products · Features · Whitepapers · Sustainability · Tokenized carbon credits · Real-world asset tokenization · SDK · Blog · Plain-text site index · Sitemap · Contact: contact@aurigraph.io