Algorithm
ML-DSA-65 (CRYSTALS-Dilithium)
ML-DSA-65 · FIPS 204 · Patent Pending · Filed 2026-05-08
smshPQ™ · What it is, and why now
Plan today's evidence for post-quantum verification.
smshPQ is a pilot integration for ML-DSA-65 signed statements. FIPS 204 defines the algorithm; a deployed receipt still needs the right schema, trusted key and successful verification. A signature records what the signer asserted. It does not establish which model executed, encrypt the contents or guarantee capture of every call.
01 · What it is
A full signature, with the material needed to check it.
The reviewed typed signer binds base envelope fields, a selected-fragment aggregate root and a policy digest into a signed 32-byte digest. The full ML-DSA-65 signature is carried inline, not replaced by a compact reference. Verification also needs canonical fragment inputs and an independently trusted public key. The three retained operator envelopes passed local cryptographic checks on 2026-09-05; production integration acceptance and underlying business claims remain separate.
Standard
FIPS 204 · Finalized Aug 2024
Integration
Pilot; endpoint acceptance required
Raw signature
3,309 bytes · 4,412 base64 characters
Illustrative fee
$0.00018 per receipt
Audit lifetime
Preservation and migration dependent
Reference: NIST FIPS 204, Table 2. Complete receipts are larger than signatures. See measured artifact and transport sizes and endpoint capability evidence.
The distinction
A sufficiently capable quantum computer could enable forgeries under classical signature schemes. That is an authenticity risk, not decryption of a signature. Harvest-now-decrypt-later concerns encrypted content and key establishment. ML-DSA addresses signatures; encryption and key management need their own migration plan.
02 · Why now · three planning questions
Choose the algorithm, evidence and retention together.
Aug 13, 2024 · Standard published
FIPS 204 finalized
NIST published the Module-Lattice-Based Digital Signature Standard, including ML-DSA-65. Using a standardized algorithm is distinct from implementation testing, algorithm validation, module validation and deployment acceptance.
csrc.nist.gov/pubs/fips/204/finalRequirements · Deployment specific
Match the required parameter set
National-security procurement rules have their own algorithms, parameter sets and transition schedules. An ML-DSA-65 label alone does not establish CNSA 2.0 compliance. Check the applicable profile before selecting an implementation.
media.defense.gov · CNSA 2.0Audit requirements · Review applicability
Logging is more than signing
Evaluate applicable AI logging duties with the responsible legal and compliance reviewers. A signed receipt can support an evidence process; it does not establish compliance with Articles 12, 13 or 50, regulatory approval or an approved retention schedule.
artificialintelligenceact.eu · Article 12Set requirements per workflow. Confirm capture coverage, accepted issuer keys, supported verifier formats and a tested retention and recovery plan.
03 · The PQ gap
Ed25519 / ECDSA versus ML-DSA-65, on what matters for an audit log.
| Dimension | Ed25519 / ECDSA | ML-DSA-65 (smshPQ) |
|---|---|---|
| Signature scheme family | Elliptic-curve, pre-quantum | Module-lattice, post-quantum |
| NIST FIPS status | Classical signature standards | FIPS 204 · Finalized Aug 2024 |
| Deployment acceptance | Endpoint, schema and key specific | Endpoint, schema and key specific |
| Quantum forgery posture | Vulnerable to a sufficiently capable quantum computer | Designed for post-quantum resistance |
| Content confidentiality | Separate encryption required | Separate encryption required |
| Sign latency | Implementation and hardware dependent | No accepted deployment benchmark |
| Raw signature size | Ed25519: 64 B; ECDSA depends on curve and encoding | 3,309 B; complete receipt is larger |
| Audit-log lifetime | Preservation, trust and migration dependent | Preservation, trust and migration dependent |
ML-DSA-65 signatures occupy 4,412 characters in base64 before surrounding JSON fields and keys. No universal expiration year applies to classical records, and post-quantum signatures do not make evidence permanent.
04 · Why not wait
Preserve the evidence needed for future checks.
Keep verification material.
Preserve signed bytes, canonicalization rules, algorithm context, trusted key history and any required fragment or inclusion evidence. A missing signature or an unavailable key can prevent verification regardless of the algorithm's strength.
Test migration and recovery.
Choose a deployment-specific retention schedule and test backup restoration and historical verification after key rotation. Crypto-agility metadata records a plan; it does not execute a migration or recover a lost artifact.
Hive endpoint capability evidence05 · The economics
Test economics against your own workload.
The calculator is a scenario model, not a customer benchmark. Enter call volume, token price, eligible cache share, routing share after caching, cheaper-model discount, receipt fee and other operating costs. It reports negative net benefits as costs and does not assume regulatory approval, utilization gains or guaranteed payback.
Illustrative fee
$0.00018
Caching
Eligibility measured separately
Routing
After cache hits only
Operating costs
Include serving and retries
Net benefit
May be negative
Commercial terms
Confirm in pilot scope
Scope a pilot.
Bring call volume, model mix and evidence requirements. Define the integration and acceptance tests before committing production traffic. Technical claims reviewed 2026-09-05.
[email protected]