One operating system. Three levels of control over the data.
The market for legal AI is dividing by where the material is allowed to be processed. Bernio is the operating system that answers all three ways — hosted, in a firm’s own infrastructure, or on a dedicated appliance.
Three things changed inside eighteen months.
A court just priced the risk
In February 2026 the Southern District of New York held that material generated through a consumer-tier AI assistant was protected by neither privilege nor work product. Until then, "do not put privileged material into a chatbot" was a policy preference. It is now a citation, and every general counsel has been handed the memo that justifies a budget line.
United States v. Heppner, S.D.N.Y., 10 Feb 2026 →European rules stopped being theoretical
Article 50 transparency obligations under the AI Act became applicable on 2 August 2026, and the deferral of the high-risk regime to December 2027 did not touch them. Combined with professional-secrecy law in civil-law jurisdictions, control over where inference runs has moved from a premium feature to a precondition for a class of work.
Our AI Act position →The capability finally clears the bar
Small models are now good enough on structured legal tasks — extraction, coding, chronology, first drafts — to run the majority of matter work on hardware a firm can own, with frontier models routed to only where policy permits. That was not true two years ago, and it is what makes the architecture commercially viable rather than merely principled.
Why the obvious acquirer does not simply do this.
One system, three deployments
A cloud-native platform is built for a single deployment model, and its unit economics, release cadence and support model all assume it. Bernio runs the same matter model, approval gates, model routing and signed record hosted by us, inside a customer’s own infrastructure, or on a dedicated appliance. We compete for hosted work on its own terms, and for the work that cannot be hosted without needing a second product to do it.
The domain depth is hard to fake
Privilege as a property of a document rather than a filter; families kept together through review; date precision carried into the argument; agent time that can never become an attorney hour. These come from practice, not from research, and they are what a firm notices in the first demo.
The evidence layer compounds
A signed, append-only record of every action is simultaneously the defensibility artefact, the billing narrative source and the answer to a client’s AI questionnaire. One artefact, three buyers, and it gets more valuable the longer a firm runs on it.
It sells alongside, not against
A displacement sale against an entrenched platform is not winnable at this stage. A complement sale into the workloads that platform cannot serve is — which is a materially shorter path to a first reference customer.
This problem needs two kinds of depth. The team has both.
One half is the obligation itself. Dr. Sergey Milutin practised for many years at Hogan Lovells in Germany, advising on cross-border disputes, investigations and compliance, and the team includes practising advocates who carry the same duty today. Requirements written by people who have taken a privilege call under deadline are not the requirements that come out of customer interviews.
The other half is research. Dr. Alexandra Carl Bernadotte works across machine learning, mathematics and robotics, with a record of shipping models under hardware constraints in fields where a wrong answer is a critical event rather than a bad draft; alongside her are systems and security engineers who have built for constrained environments before. Running capable models inside a controlled perimeter, with per-matter isolation and a signed record, is a research and systems problem before it is a legal one — and it is the half that decides whether the legal half is deliverable at all.
Status, stated plainly: the platform is in development with design partners — firms running real matters through it and shaping what gets built next — and the first commercial contracts are being signed on founding terms. Bernio Pro, the appliance that carries the strongest form of the isolation argument, is an engineering prototype and is not yet commercially available.
The briefing
Round size, instrument, valuation, use of proceeds, financial model, pipeline, go-to-market sequence and roadmap dates are in the investor briefing, shared under NDA. We respond within one business day.
Request the investor briefing.
We will share the current materials and development status under NDA.