Deterministic AI · Not an LLM

Building superintelligence humanity can trust.

Proven — not taken on faith

The Platform

What is Sentium X?

A new class of AI, engineered for institutions that cannot afford black boxes. Plug it into your existing systems and get decisions that are exact, explainable, and provable.

Every AI can explain itself. None can prove the explanation is true.

Ask an AI why it decided something and it writes the answer afterwards — using the same machinery that made the decision, with no obligation to accuracy and nothing to check it against. Ask twice and you may get two different answers.

That is a witness statement: produced after the event, by the party involved, unverifiable by anyone else.

We do not explain the decision. We re-run it. Same inputs, same result, to the last digit, on any machine — checked by whoever needs convincing, with no access to our systems and no need to take our word for anything.

Deterministic by design

The same input produces the same output — no randomness, no drift — so decisions can be reproduced and reviewed rather than taken on trust. Measured: 300,000 computation steps re-run byte-identically across six independent runs.

Not a language model

Sentium X is not an LLM. Rather than generating probabilistic guesses, it computes decisions through traceable, deterministic logic — built for environments where every decision has to be explained.

Transparent to its core

Every computation is recorded in a tamper-evident ledger from input to output — so regulators, auditors, and your own teams can inspect exactly why any decision was made.

The memory problem

Every AI in production deletes its own past

Not by choice. By arithmetic.

Every transformer pays quadratically to remember: ten times the context, up to a hundred times the compute. Cached instead, a single million-token conversation needs 125 GB — more than any single GPU holds. Time or memory. There is no third option.

So they forget. Sliding-window attention — the standard workaround — keeps one seventh of its accuracy at 128,000 tokens. A model advertising 262,000 tokens of context delivers an effective 32,000.

We had the same disease and measured it in our own engine: ten times the history cost forty-four times the memory. It now costs 8.7. Nothing summarised, nothing evicted, nothing approximated — and every decision still replays to the bit.

100×

ten times the data costs them a hundred times more

quadratic attention, peer-reviewed
1/7

the accuracy left after they forget to cope

Microsoft Research, RULER @128K
8.7×

ten times the data costs us 8.7 times more

measured, six seeds, n=3,000
894 → 125 MB

our complete history, output byte-identical

replay it yourself

Ten times the data costs them a hundred times more.

On our substrate, ten times the data costs 8.7 times more.

That gap is the $725 billion.

Forgetting is the industry standard. We broke it — and kept the receipts.

Liu et al., RetrievalAttention, Microsoft Research, arXiv:2409.10516, Tables 1 and 3. The 44× and 8.7× figures are our own engine, measured under a pre-registered specification.

Key Benefits

Precision you can take to the regulator

Built from first principles for accountability — so trust becomes a property of the system, not a promise.

Full audit trail

Every computation is logged end-to-end, so decisions can be reconstructed and independently checked after the fact.

Complete transparency

No black boxes. The full reasoning path behind every output is open to inspection by your teams and regulators.

Input-to-output ledger

A continuous, tamper-evident record links every input to its output — a chain of custody for your decisions.

Runs where the data is

No specialised accelerators, no cloud lock-in. And because the history is held efficiently rather than discarded, the whole record stays on the machine that made the decisions.

Compliance-ready

Explainability, traceability, and human oversight are native to the architecture — not bolted on afterwards.

Deploy inside your walls

Install on-premise within sensitive environments. Your data and your decisions never have to leave your perimeter.

Interference changes the answer

When identical inputs must produce identical outputs, any difference is itself an alarm. Model poisoning, altered data and supply-chain compromise become detectable — not because we added a detector, but because there is nothing for them to hide in.

Industries

Built for the institutions the world depends on

Wherever decisions carry regulatory, financial, or human consequence, Sentium X plugs in and stands up to scrutiny.

Accountable by mandate

Government

Public-sector decisions demand public accountability. Deliver services and determinations citizens can verify — and oversight bodies can audit line by line.

Safety-critical

Healthcare

When outcomes affect lives, “probably right” is not enough. Sentium X is built to support clinical and operational decisions with reproducible logic and a full evidentiary record.

Examination-ready

Banking & Finance

Credit, risk, and compliance decisions with a complete paper trail. Answer any regulator, resolve any dispute, and prove fairness on demand.

Certifiable control

Critical Infrastructure

Power grids, water systems, transport networks. Deterministic control logic that behaves identically under pressure — and can be certified before deployment.

Regulation

Engineered for the era of AI regulation

The EU AI Act (Regulation 2024/1689) requires high-risk AI systems to be transparent, traceable, and subject to human oversight. Most AI retrofits compliance. Sentium X produces the evidence compliance requires, as a by-product of how it computes.

The EU AI Act changes everything

Under Article 6 and Annex III, AI systems in healthcare, finance, government services, and critical infrastructure are classified as high-risk — and must prove how they work. Opaque models face costly conformity assessments, documentation burdens, and Article 99 penalties of up to €35 million or 7% of global turnover.

Sentium X is designed to turn that burden into an advantage: the evidence regulators require is generated as a by-product of how the system computes.

Regulation (EU) 2024/1689High-risk: 2 Dec 2027
inputdeterministic logic output
└ ledger: complete · verifiable · tamper-evident
Article 13

Transparency & explainability

Every output ships with its complete reasoning path — readable by humans, verifiable by machines. Article 13 requires high-risk AI systems to be transparent enough for deployers to interpret and use their output.

Article 12

Traceability & record-keeping

The input-to-output ledger captures the automatic logging and record-keeping obligations Article 12 places on high-risk AI systems, out of the box.

Article 14

Human oversight

Article 14 requires high-risk AI to allow effective human oversight. Deterministic behavior makes meaningful human review possible — reviewers see exactly what the system will do, and why.

Article 9

Risk management & robustness

Article 9 demands a continuous risk management system across the lifecycle. Reproducible outputs mean testable systems — validate once, and the behaviour holds, verified across machines and across a compiler change, in audit and in production.

€35M / 7%

maximum Article 99 penalty — €35M or 7% of global turnover for prohibited practices; €15M or 3% for high-risk violations

2 Dec 2027

high-risk enforcement date — moved from Aug 2026 by the Digital Omnibus, adopted June 2026

$5.78B by 2029

AI governance market, up from $0.89B in 2024 — a 45.3% CAGR (MarketsandMarkets, 2024)

$725B

annual AI infrastructure spend by the four largest technology companies, 2026 — up ~77% year on year (company capex guidance, 2026)

Beyond the EU: in April 2026, US banking regulators — the OCC, Federal Reserve and FDIC — issued revised model risk management guidance. It is principles-based rather than mandatory and does not yet cover generative or agentic AI, but for the models in its scope it treats reproducibility as central to validation, expecting deterministic models to produce identical outputs for identical inputs.

OCC Bulletin 2026-13, April 2026

The Founder

Built by someone who has to stand behind it

Sentium X is the work of a founder building the accountability layer that regulated institutions have been asking AI for.

LM

Dr Luke McCabe

Founder

LinkedIn

He started Sentium X because the institutions that can least afford a wrong answer were being asked to trust AI they could not audit.

Luke spent two decades in clinical practice — running a healthcare business in a regulated, high-accountability environment where a decision is never judged only on whether it looked right, but on whether you can show exactly how it was reached and stand behind it. That is the discipline of any regulated field: the record matters as much as the result.

He watched AI move into healthcare, finance, and public services carrying none of that discipline — systems that produce an output but cannot show their work. Sentium X is his answer: an AI built so that every decision is deterministic, reproducible, and recorded from input to output, so the people accountable for a decision can prove how it was made — not just trust that it was right.

Eighteen months of engineering, delivered self-funded, producing a core whose reproducibility is machine-verified rather than asserted — including a memory result measured under acceptance bars written before the measurement existed.

The Reproducibility Audit

Start by checking us, not buying from us

The fastest way to trust a system that promises reproducibility is to reproduce it yourself. So that is where we start.

01

Send one month of decision logs

No integration, no procurement, no fee. Just a month of the decisions you already make.

02

Four weeks later, you have a finding

Your own assessors establish whether those decisions reproduce — independently of us.

03

The deliverable is their finding, not our claim

You start by checking us, not by buying from us. The proof belongs to you.

No fee. No integration. No procurement.

Early Access

Be first in line for provable AI

We are onboarding a limited number of partners in regulated sectors. Join the waitlist to secure early access to Sentium X.

Your email is stored securely and used only to contact you about Sentium X.

Contact

Talk to the team

Evaluating Sentium X for your organization, or exploring a partnership? Send us a message.

Your details are stored securely and used only to respond to your inquiry.