Post-Quantum Inference · ML-DSA-65 · FIPS 204 · Patent Pending · Filed 2026-05-08

The only inference rail that's already post-quantum.

smshPQTM signs every inference call with a NIST-finalized lattice signature in under a millisecond. Today's classical signatures become forgeable the day a cryptographically-relevant quantum computer arrives. Every audit log, every routing decision, every cache hit you sign today with Ed25519 or ECDSA is harvested now and decrypted later. smshPQ closes that window — and pays for itself on the cost side while it does.

Per receipt$0.00018
·
Sign latency<1 ms
·
SchemeML-DSA-65
·
AnchorBase
·
Receipt size384 B
Drop-in inline with OpenAI Anthropic Google Kimi Manus vLLM SGLang Self-hosted
The quantum window · Open today · Closes the day a CRQC ships

Every inference signed with classical curves is a future liability.

NIST finalized ML-DSA in FIPS 204 in August 2024. The NSA put it on the CNSA 2.0 list with a 2035 hard deadline for national-security systems. The harvest-now-decrypt-later attack does not need a quantum computer to start — adversaries are storing your signed traffic right now. Inference receipts are the longest-lived secrets you produce.

Property
smshPQTM
Classical-signed railsEd25519 / ECDSA
Unsigned inferenceToday's default
Signature scheme
ML-DSA-65 (lattice)
Ed25519 / ECDSA
None
NIST FIPS status
FIPS 204 finalized · Aug 2024
Pre-quantum standards only
n/a
NSA CNSA 2.0 listed
Yes
No · deprecation by 2035
No
Quantum-forgeable
No
Yes · Shor breaks it
n/a · nothing to forge
Harvest-now-decrypt-later exposure
Zero
Full · adversary stores now, breaks later
Full · no integrity at all
Sign latency
<1 ms
~0.2 ms (faster, less safe)
0 ms · no proof
Receipt size
384 bytes
64 bytes
n/a
Audit-log lifetime
Unbounded · post-quantum permanent
~5 years · CRQC dependent
Zero · cannot be audited
The lead
smshPQ is the only inference primitive shipping today on a NIST-finalized post-quantum signature. The cost lever pays for the upgrade. The PQ posture is the moat. By the time the rest of the rail catches up, the audit log you signed in 2026 is the only one that still verifies in 2035.
Calculator · Your workload, your numbers

Your current inference spend vs. your spend with smshPQ™.

Enter your monthly call volume and current cost. We'll show the cache-proof savings, the verified-routing savings, and the net cost after the smshPQ™ receipt fee. No accounts. No gimmicks. The math is the math.

Net annual savings
ROI on smshPQ™ fee
Payback
Monthly receipts
Pick your scale
your number, or pick a scale above
e.g. 800 in + 400 out
e.g. $3.50 blended
repeat prompts you currently re-run · typical 15-30%
prompts a smaller proven model can answer · typical 30-45%
determines deployment unlock
Current state Today
Monthly tokens
Monthly inference spend
Cache-proof savings$0
Verified-routing savings$0
Audit ceiling
Annual all-in cost · today
With smshPQ™ Signed inference
Monthly tokens
Monthly inference spend
Cache-proof savings
Verified-routing savings
Audit ceiling100%
smshPQ™ receipt fee
Annual all-in cost · with smshPQ™
Net annual savings: · ROI · Payback in
Three savings levers · One primitive

Why the math always works.

A signed hash is small, fast, and proves three things at once: which model ran, what scope it was allowed, and whether the answer was already paid for. Each unlocks a different line of cost.

01 · Cache-proof savings

Prove a result was already paid for.

If a prompt+context hash matches a prior signed response from the same model, serve it from cache with cryptographic proof of equivalence. The buyer trusts the hit. The vendor saves the inference.

Typical recapture: 15-30% of repeat traffic
02 · Verified routing savings

Route to the cheaper model — and prove which model answered.

Today, buyers overspend on premium models because they cannot verify which model actually answered. smshPQ™ signs the model ID into every receipt. Route freely between Claude Haiku, GPT-4.1-mini, Gemini Flash — and prove it on the way out.

Typical recapture: 30-50% of routable traffic
03 · Audit unlock

Deploy at full capacity in regulated workloads.

Most regulated buyers run inference at 30-40% of capacity because they cannot audit at scale. smshPQ™ gives the regulator the receipt, the counterparty the proof, and the holder the keys — without revealing the prompt or the output.

Unlocks the other 60-70% of capacity
How it works · 4 steps · <1ms

Call. Hash. Sign. Anchor. Every inference, every time.

smshPQ™ sits inline with the inference call. Zero model change. Zero protocol break. The receipt is a 384-byte ML-DSA-65 signature over a canonical hash of (prompt, model, scope, output) — anchored on Base in batched commits.

01 · Call
Inference runs
Any model, any provider. OpenAI, Anthropic, Google, Kimi, Manus, or self-hosted. Zero change to the model call.
02 · Hash
Canonical SmSH
A canonical small-message signed-hash binds (prompt, model, scope, output) into one 32-byte digest. Byte-match validated.
03 · Sign
ML-DSA-65 PQ
A post-quantum lattice signature is generated in under a millisecond. The receipt travels with the response.
04 · Anchor
Base commit
Receipts are batched and anchored on Base. Any party — regulator, counterparty, holder — verifies independently.
Pilot scoping · 20 minutes · NDA available

Bring us your call volume. We'll bring the math.

Send a single line of API traffic. We'll instrument smshPQ™ inline, show your real cache-proof rate and verified-routing rate against your actual prompts, and walk you through the deployment in a 20-minute call.

Book the pilot scoping call
smshPQ™ · Patent Pending · Filed 2026-05-08 · Anchored on Base