The Legal Studio: Privilege Safe AI for Law Firms

Privilege is fragile, and cloud AI treats it carelessly

Legal professional privilege is one of the oldest protections a client has. It lets a person speak candidly to a solicitor, knowing the advice and the confidences behind it cannot be dragged into the open by an opponent or a regulator. That protection is powerful, but it is also fragile. Privilege can be waived, and once it is waived it is very hard to claw back. The classic way to lose it is to share the privileged material with a third party who has no need to see it. When that happens, the shield can drop for the whole document, and sometimes for related documents too.

This is exactly where mainstream cloud artificial intelligence becomes a quiet liability for a law firm. When a lawyer pastes a draft advice, a witness statement, or a set of disclosure documents into a hosted model, that material travels to servers the firm does not own, controlled by a company the client never retained. The text is processed, often logged, sometimes retained for a period, and in some arrangements used to improve the service. Each of those steps is a disclosure to a third party. A court will not care that the disclosure felt convenient. It will ask a simple question. Did the privileged content leave the circle of confidentiality? With most cloud AI, the honest answer is yes.

We think the profession deserves better than a choice between falling behind on technology and quietly eroding the very protection that makes candid legal advice possible. That is why we built the Legal Studio inside Mickai, and why we built it on a foundation that never sends privileged work anywhere at all.

What Mickai is, and why the architecture matters here

Mickai is a Sovereign Intelligence Operating System, a SIOS, held privately by its founder. The word sovereign is not decoration. It describes where the intelligence actually runs. Mickai runs on the customer's own hardware, on premises, and it can run fully air gapped with no connection to the public internet. There is zero data egress and no public cloud round trip. The firm's documents are read, reasoned over, and drafted against on machines the firm physically controls, sitting behind the firm's own locks and its own network boundary.

For a law firm, that single design decision changes the privilege analysis completely. If the privileged material never leaves the firm's control, there is no disclosure to a third party, and so there is nothing to waive. The confidence stays inside the circle. The intelligence comes to the documents rather than the documents being shipped out to the intelligence.

Under the surface, the Legal Studio draws on a wider system of 50 specialist brains, 25 domain brains and 25 operational brains, coordinated under deterministic governance. Instead of one general model doing everything, the work is routed to the specialists suited to it, and their behaviour is constrained by rules rather than left to chance. For legal work that discipline matters, because a firm needs to be able to say not just what the answer was, but why the system was allowed to produce it.

The proof a firm can hand to a regulator

Keeping data in the building is necessary, but on its own it is not enough. A firm also has to be able to demonstrate what happened. Auditors, regulators, professional indemnity insurers, and opposing parties in litigation may all one day ask how a piece of AI assisted work was produced. Mickai is built to answer that question with evidence rather than assurances.

Every action the system takes is written to a cryptographically signed audit record, the Open Audit Record. That means each step, from the moment a document is read to the moment a draft is produced, leaves a tamper evident trace. The signatures use post-quantum signing, specifically ML-DSA-65, so the record is designed to remain trustworthy even as cryptography moves into the post-quantum era. And the memory the system builds up over time, the accumulated knowledge of a matter, belongs to the customer. It is not a shared pool feeding someone else's model. It is the firm's own asset, held on the firm's own storage.

In practical terms, a firm using the Legal Studio can show the following, and show it with proof rather than a policy document:

• Privileged material stayed on the firm's own hardware and never crossed the network boundary, so it was never disclosed to a third party.
• Every AI assisted action carries a cryptographically signed entry in the Open Audit Record, giving a tamper evident trail from source document to final draft.
• The signing is post-quantum, using ML-DSA-65, so the evidence is built to hold up over the long life of a legal matter.
• The knowledge the system accumulates about a matter is memory the firm owns, not a resource harvested by an outside operator.
• The specialist brains work under deterministic governance, so the firm can explain the boundaries the system operated within.

“The safest way to keep privilege intact is simple. Never let the privileged material leave the room. We designed the Legal Studio so that the intelligence comes to the documents, and the documents never have to go anywhere.”

What the Legal Studio does with all of that

The point of privilege safe architecture is to free lawyers to actually use the technology rather than tiptoe around it. Because the material stays inside the firm, the work that was previously too sensitive to hand to a hosted model becomes fair game. A team can run first pass review across a disclosure set, surface the documents that matter, and keep the privileged and the merely confidential clearly separated. Associates can draft against the firm's own precedents and prior advice, with the system reasoning over the firm's real knowledge rather than a generic public corpus. Fee earners can interrogate a bundle, test an argument, and prepare for a hearing, all without a single privileged sentence leaving the building.

Because the memory belongs to the firm, the system also gets more useful the longer it is used. It learns the house style, the standard positions, the way a particular partner likes advice framed, and it keeps that knowledge in house. None of it leaks out to sharpen a competitor's tool. The value compounds, and it compounds for the firm that created it.

Why this matters now

The pressure on firms to adopt AI is only going to grow. Clients expect faster turnarounds and keener pricing, and both push towards automation. But the profession also carries duties that predate every model on the market, and confidentiality sits at the centre of them. A firm that solves the speed problem by quietly waiving privilege has not solved its problem at all. It has simply moved the risk somewhere it will not be noticed until it is too late.

We are not asking the market to take our ambition on faith. Our own public signal has been building in the open. Our founder now ranks number 2 on Crunchbase, and the company Heat Score reached 94 out of 100, climbing from single digits. That is a public, independently visible marker of momentum, and we would rather point to it than to claims that cannot be checked.

The intellectual property behind the system is being protected with the same seriousness. We have 104 filed UK patent applications, containing approximately 2,340 claims across their full specifications, claims, and figures, and we are building steadily towards examination and grant. The architecture that keeps privilege intact is not an afterthought bolted onto a cloud product. It is the thing we set out to build.

Looking ahead, we expect privilege safe, on premises intelligence to become the baseline expectation for serious legal work rather than a premium option. The firms that move early will have learned how to fold sovereign AI into their practice while their confidences stayed exactly where they belong. We built the Legal Studio for those firms, and we are ready to help them make the move on their terms, on their hardware, with the proof in their hands.