About
We build AI systems that have to show their work.
Kenshiki Labs builds the runtime control and proof layer for consequential AI decisions. The founders bring four decades each of relevant experience: Stephen Fishburn from statistical process control with Deming and Juran's lineage at Japan Management Association Consultants, Bell Labs research, and 120M-subscriber pattern-of-life infrastructure; Larry Signorile from Battelle, Classmates.com platform re-engineering, and 17 years at CBC Innovis architecting high-throughput regulated-data APIs at 2,000 TPS. The team chose runtime AI governance because the problem requires production-systems discipline applied to AI inference, not the other way around.
What Kenshiki Labs is
Kenshiki Labs is an AI governance control plane. The Prompt Sanitizer establishes caller identity. Kura stores governed evidence with provenance, SIRE identity, and retrieval boundaries. The Compiler assembles governed prompts from four CFPO zones. Kadai generates bounded answers from that evidence. The Gate checks every claim against the evidence before emission. The Ledger records a tamper-evident chain of custody for every response. Two APIs. One contract. Same system whether you run on shared cloud or inside an air gap.
- Prompt Sanitizer → Kura → Compiler → Kadai → Gate → Ledger
- Kura stores governed source material with provenance, SIRE identity, and retrieval boundaries
- Kadai returns answers bounded by what Kura can support
- The Claim Ledger and Gate evaluate what can actually be emitted
What Kenshiki Labs is not
The wrong mental model slows people down. Kenshiki Labs is not another model endpoint, not a content-moderation layer, and not a monitoring dashboard that tells you after the fact that something went wrong. It is the system that governs what the generation layer can rely on and what it is allowed to emit.
- Not a model — it governs the generation layer instead of replacing it
- Not a content filter — it checks evidence, not tone or topic
- Not a monitoring tool — it intervenes before emission, not after reliance
- Not a GRC dashboard — it enforces policy at runtime, not as a documentation exercise
Why Kenshiki Labs exists
AI rarely fails like normal software. There is no obvious crash, no red error state, no stack trace for the person relying on the answer. It fails by sounding authoritative before anyone has established whether the authority is real. Kenshiki Labs exists because bigger models do not solve that problem. Better fluency only makes unsupported reasoning harder to spot.
What makes Kenshiki Labs different
We are not trying to make a model look more responsible. We are moving authority out of the model entirely. The model can propose language. It cannot establish its own authority, invent its own provenance, or decide on its own that an answer is safe to rely on. Authority lives in the evidence, the policy, and the gate — not in the weights.
- No governed evidence, no decision-grade emission
- Authority lives outside the model
- Every consequential answer is reduced to claims that can be checked
- The same governance contract holds across Workshop, Refinery, and Clean Room
How it runs
Kenshiki Labs is designed so teams can start where they are and deepen the assurance boundary over time. Workshop lets you use shared Kadai or the public-model APIs you already have. Refinery moves the same system into your environment with stronger local control. Clean Room runs the full stack inside an air-gapped boundary when external connectivity is not an option.
- Workshop: shared Kadai with governed retrieval and claim checking
- Refinery: private deployment inside your VPC with full telemetry and chain of custody
- Clean Room: air-gapped execution with local storage and secure media transfer — no network path
Who it is for
Kenshiki Labs is for teams operating where a fluent mistake can move money, shape care, expose intelligence, or create legal and regulatory risk. We build for operators who will be asked to show their work later, not just teams trying to ship a demo now.
- Defense and intelligence workflows where sourcing must survive review
- Government and public-sector systems that face oversight and disclosure pressure
- Healthcare and life-sciences teams that need evidence-backed recommendations
- Regulated enterprises that must explain outputs under audit, litigation, or policy review
How to evaluate us
The right way to assess Kenshiki Labs is not by whether the prose sounds good. It is by whether the system can show what evidence was in scope, what claims were made, what held up, what failed, and why an answer was allowed to leave at all. That is why we publish our architecture, pricing model, and failure cases openly.
- Read the platform architecture to see where authority actually lives
- Review pricing to understand how governance is separated from raw inference
- Use the AI Incident Archive to compare our design against real failure modes
- Try Kadai and inspect the audit trail for any response
Why the names
The names are Japanese. We use them because Japanese has single words for concepts that English needs a sentence to describe — and because the governance problem we solve is fundamentally about precision of meaning. A system that enforces the boundary between verified evidence and fluent assertion should be named in a language where that distinction already has a word.
- Kenshiki Labs (検識) — verification through inspection. The act of examining something to establish its true nature. Not trust, not belief — inspection. That is what the company does.
- Kura (蔵) — a storehouse. The traditional fireproof warehouse where valuable goods are kept safe and accounted for. Kura is the governed evidence store — the boundary that keeps evidence intact and retrievable under policy.
- Kadai (課題) — the question that must be answered. Not a casual question — a formal problem that demands a rigorous, structured response. Kadai is the governed inference engine that produces bounded answers from governed evidence.
Leadership
Kenshiki Labs was founded by operators whose combined experience spans cryptanalytic rigor, Tier 1 platform engineering, continental-scale data infrastructure, and regulated enterprise systems — the disciplines the governed-AI problem actually requires.
- Stephen Fishburn, co-founder. Stephen has spent his career at the intersection of rigorous systems thinking and production-scale data infrastructure. He began at Japan Management Association Consultants in the late 1980s, working on statistical process control and quality function deployment alongside Drucker, Deming, and Juran — the founders of modern quality management. He holds an M.S. in computer science and mathematics from the Tokyo Institute of Technology, with a thesis in structural cryptanalysis of block ciphers. He was a researcher at Bell Labs and served on Microsoft's Windows Base Team. At UIEvolution, he was instrumental in establishing the company as a Tier 1 automotive supplier to Toyota and General Motors, architecting the connected-vehicle platform behind Toyota's Entune — he is the inventor of the underlying peer communications framework (US Patent App. 2011/0126014). Most recently, as Chief Data Scientist at a geospatial audience-analytics company, he built the continuous pattern-of-life pipeline resolving home, work, and visit locations for 120 million mobile subscribers against a building-level POI database and 6,000-attribute identity graph. Bilingual in English and Japanese.
- Larry Signorile, co-founder and CTO. Larry leads Kenshiki Labs' engineering team and owns the core of the platform — Kura and Kadai. He brings nearly four decades of production software experience across regulated data systems, high-throughput transaction platforms, and enterprise security architecture. He began his career as a researcher at Battelle, where over nine years he delivered systems for the US Air Force (CPAS accounting), the US Army (bibliographic databases), state lottery acceptance testing, and secure smart-card financial kiosks — work conducted under Good Laboratory Practice certification. He went on to lead the re-engineering of Classmates.com from a collection of Perl scripts into a modern n-tier Java platform, meeting 100ms query performance targets across 40 million users, reporting to the CTO and presenting to the board. Over 17 years at CBC Innovis — a regulated consumer-data and verification business serving the mortgage and credit industry, where he reported to the owner — he advanced to VP of Software Engineering, owning enterprise application development, API architecture, and transaction processing. He architected the high-performance API for the company's primary transaction system, scaling throughput from 100,000 transactions/hour to 2,000 TPS (~7M/hour) at ~100ms response times; directed the migration from SOAP to REST; led the implementation of OAuth 2.0, encrypted payloads, and SCIM 2.0 identity interfaces; and served as the engineering organization's Security Champion. He studied computer and information sciences at The Ohio State University.
Start Here
See how the contract actually lands.
If you're evaluating fit, pricing is the fastest next read. If you want to inspect the runtime contract, go to architecture. If you want to talk through your environment, contact us directly.