A new way for industrial operators to turn hard-won experience into working AI — where everything that identifies them stays theirs and in their tenant, and the de-identified pattern comes back to them free, forever, improved by work done across four industries.
Irvine, CA — August 6, 2026 — EON AI Ventures today introduced EON Delta, working AI for industrial operations built under EON’s published three-tier data rights standard. Everything that identifies a customer stays in that customer’s tenant and remains their property. What EON carries forward is the de-identified pattern — and every participating operator receives that back, royalty-free and in perpetuity.
EON Delta is aimed at operators of complex physical infrastructure — energy, pipelines, manufacturing, transport and utilities — where general-purpose AI has so far struggled to be useful, and where sending operational data to an external cloud service is often prohibited outright. Read more in the Owning the Trained AI Model white paper.
Click on the image below to access the EON Delta presentation.
Why: two problems that have kept AI out of heavy industry
Today’s AI systems are brilliant generalists that know nothing in particular. They have never seen a specific plant, do not know its procedures, and have no way of learning whether the advice they just gave was correct. Dropped into an industrial setting, they produce fluent recommendations that no experienced engineer would act on.
The second problem is more basic. Operators of critical infrastructure frequently cannot send live operational data to a third-party cloud at all. It is not a preference — it is a condition of their licence, their insurance, or their national regulator. That single constraint has kept most commercial AI off the table before a conversation can even begin.
Meanwhile, the people who genuinely understand how these facilities behave are retiring. Their judgment was never written into any procedure document, and it cannot be interviewed out of them at scale.
| “Every industrial company is sitting on decades of hard-won experience that was never written down, and it is walking out of the door as people retire,” said Dan Lejerskar, Founder of EON Reality. “EON Delta captures that experience and hands it back to the people doing the work — without a single byte of it leaving the customer’s building.” |
What: a model in two halves
An EON Delta is a compact, specialised AI model small enough to run on a laptop, a vehicle, or a device a technician carries into the field. It is built in two layers, and the distinction between them is the whole point.
| The Tenant Delta — the customer’s half.
Trained inside the customer’s own tenant, on the customer’s own records, and never leaving it. It knows their asset tags, their sites, their local procedures. It is their property, it is returned or destroyed if they leave, and EON does not use it to build anything for anyone else. The Base Delta — EON’s half. Trained on de-identified patterns — equipment failure modes by class, precursor signatures, how people systematically deviate from procedure. No pattern enters it unless it has been seen at two or more unaffiliated organisations. This half is EON’s, and every participating operator receives it free of charge, in perpetuity, for as long as they want it. |
On the device the two run together. The Base Delta supplies judgment learned across an entire industry; the Tenant Delta supplies everything specific to the site. The customer’s half never leaves their building. EON’s half gets better every month from work performed elsewhere — at no additional cost, and without any customer’s data or identity moving anywhere.
Each Delta also ships with the verified dataset that produced it and a benchmark score measured against real history, so performance is audited rather than asserted.
How: three steps,
- Capture what the experts know. EON assembles training material from our EON Sitaution Room Incidents (EON Delta Zero) and inside the customer’s own tenant, their existing records — inspection reports, work orders, incident write-ups — together with scenarios generated inside EON’s digital twin technology. Every example is checked by a qualified engineer. This material does not leave the tenant.
- Train a small, specialised model. Rather than building an AI from scratch, EON adapts an existing US aproved open model by training a small additional component on the verified material. This takes hours rather than months, and costs a fraction of one percent of building a model from nothing.
- Deploy it where the work happens. The finished model runs inside the customer’s own perimeter — including fully offline, on a portable device with no network connection. When it is unsure, it says so and defers to the technician. Every correction the technician makes is captured and folded into the next version.
That last point is what makes the approach compound. The system does not simply hold its performance — it improves with every shift worked, using knowledge that no competitor can obtain.
What makes it different
| The AI never touches the pipeline. It touches the twin of the pipeline.
EON Delta is strictly advisory. It reads data and produces recommendations, reports and ranked warnings for people. It is never connected to control systems, and no output of it can actuate equipment. Safety-critical control remains where it belongs: in deterministic, certified systems. |
- Data sovereignty by design — raw operational data stays in the customer’s tenant and is not used to train anything made available to any other party. This is written into the contract and is auditable, not asserted.
- Built on twenty-five years of digital twins — EON has been building simulations of physical work since long before generative AI existed. Those simulations are what allow an AI’s suggestions to be tested safely before a human ever sees them.
- Audited, not asserted — every EON Delta ships with a score measured against the customer’s own documented history, produced by a test written before the model was built.
- Reciprocity — every participating operator receives the full body of generalised patterns back, royalty-free and in perpetuity, including patterns contributed by others and drawn from every sector EON serves. On day one a joining operator is a net receiver.
- Competitively sensitive information is excluded categorically — production volumes, unit costs, yields, formulations, subsurface data, turnaround schedules, capital plans, vendor terms. Excluded by definition, not by EON’s discretion.
Data rights
EON Delta is delivered under EON’s published three-tier data rights standard.
- All raw operational data and all customer-specific analysis remain the exclusive property of the customer and stay within the customer’s tenant; they are not used to train any model made available to any other party.
- EON retains only de-identified, non-attributable patterns, and only those required to safeguard personnel, maintain equipment reliability and build workforce competency — and only where a pattern has been independently observed at two or more unaffiliated organisations, falls outside a defined list of competitively sensitive categories, is drawn from historical rather than current operations, and has passed a thirty-day customer review with a right to withhold.
Where a finding bears on life safety, neither party may suppress its disclosure.
The standard is published, contractual and auditable. EON does not sell the underlying body of patterns as a dataset to anyone; it is delivered only embedded in products that make work safer.
Read more in the Owning the Trained AI Model white paper.
Learn more by tuning to our podcast.
About EON Reality
EON AI Ventures is a knowledge-transfer company specialising in immersive and simulation-based training for industry and education. Founded by Dan Lejerskar, EON has spent more than two decades building digital twins of equipment, procedures and working environments used by industrial, government and academic organisations worldwide. EON AI Ventures extends that work into applied artificial intelligence for physical industry.