Legal work deserves infrastructure built to its own standard.
Founded in 2026 by practising lawyers, engineers and security specialists who had spent their careers on opposite sides of the same problem.
To make the most capable AI usable for the most confidential work in the economy — without a firm having to hand over control of privileged material to get it.
Every other answer to that problem is a promise: a clause in a processing agreement, an assurance that inputs are not retained, a certificate that the vendor behaved as described. Promises are worth exactly as much as the counterparty behind them, and they are tested only after something has gone wrong.
We would rather the guarantee be a property of the system. Matter isolation enforced beneath the application, no training on client data, and an audit trail signed and held by the firm — whether Bernio runs in our cloud or on your own hardware. Bernio Pro removes the network path altogether. That is an engineering problem before it is a legal one, which is why this company has scientists in it as well as lawyers.
The Bernio Core Team
Two professions that were not talking to each other.
Legal software has largely been built by people who have never reviewed a contract against a deadline or taken a privilege call with a client on the line. It shows — in tools that treat a matter as a conversation, an approval as a habit rather than a state, and an audit trail as telemetry rather than evidence.
The hard parts, meanwhile, are not legal problems at all. Running capable models inside a controlled perimeter, on constrained hardware, with cryptographic accountability, is systems engineering — and it is solved in a different building.
Bernio exists because its founders came from both buildings. Dr. Sergey Milutin practised for many years at Hogan Lovells in Germany, advising on cross-border disputes, investigations and compliance. Dr. Alexandra Carl Bernadotte is a deep-tech founder and scientist working across AI, mathematics and robotics, with a record of shipping machine learning under hardware constraints in fields where a wrong answer is a critical event rather than a bad draft.
Behind them is a team: practising advocates, systems engineers and security specialists, each brought in for the part of the problem they have solved before. Most of them do not appear on this page, which is a decision about the page rather than about the work.
Four commitments, and what they cost us.
Architecture over assurance
If a guarantee depends on us behaving well, it is not a guarantee. We say plainly which is which.
The practitioner decides
Nothing reaches a client without a qualified person putting their name to it. Automation earns scope by being auditable.
Evidence, not telemetry
The record a matter produces should be usable in the proceeding it came from.
Built with practitioners
Every workflow was drawn from how the work is actually done, and revised by people who have done it under deadline.
Built for jurisdictions where the rules decide where data may sit.
Bernio is built first for the United States, the European Union — the DACH countries in particular — the United Kingdom and Latin America. In each of them, professional secrecy, data protection and now AI regulation converge on the same question, and it is not which model a firm uses. It is where the material may be processed, and what the firm has to be able to show afterwards.
A matter touching three jurisdictions has to satisfy three sets of confidentiality rules at once, and one processing agreement does not settle them all cleanly. That is why the isolation is architectural rather than contractual: what answers the question is the tier a firm deploys on, not the clause it negotiates.
Primary markets
- United States
- European Union · DACH
- United Kingdom
- Latin America
We would rather show you than tell you.
Early access is limited to qualifying law firms and legal departments. We run the demo on a matter type your team handles.