AI in Court: What the SRA Referral Means for Verifiable Legal AI

The only defensible way for a law firm to use artificial intelligence is to verify every output before relying on it, and to hold evidence of that verification, because the courts will now refer solicitors to their regulator when AI-influenced statements reach the record unchecked. In a ruling handed down in May 2026, the UK High Court raised concerns about misleading, AI-influenced statements in letters dated 30 March and 14 April 2026, found a prima facie breach of the duty not to mislead the court, and referred the solicitor concerned to the Solicitors Regulation Authority. The duties have not changed, but the ways of breaching them have multiplied.

We read this ruling with sympathy as well as concern. Nobody sets out to mislead a court. In practice, a fluent output from a system nobody in the firm controls looks right, deadlines press, and it passes into correspondence with no record of what was generated, what was checked and who signed it off. The High Court has now confirmed where that quiet failure ends up, in front of the regulator, attached to an individual's name.

What did the High Court actually find in May 2026?

The court found a prima facie breach of the duty not to mislead, arising from AI-influenced statements in letters dated 30 March and 14 April 2026, and referred the solicitor to the Solicitors Regulation Authority. What matters for every other firm is the shape of the finding rather than the identity of the parties. The court did not need to rule on whether artificial intelligence belongs in legal practice. It applied a duty that predates the technology by centuries to correspondence carrying assertions no one had verified, and responsibility sat where it always sits, with the professional who put their name to the words.

Why is unverified AI a conduct problem rather than a technology problem?

Because the duty engaged is personal and the evidence trail usually is not. A solicitor who relies on an unverified output has delegated part of their professional judgement to a system that keeps its workings elsewhere, if it keeps them at all. When the regulator later asks what the system was asked, what it returned and who reviewed it, most firms can only reconstruct an answer from memory and fragments. That gap between what happened and what can be proved is where careers are damaged. The technology did not create the duty. It created a new and largely unrecorded way to breach it.

“A court will not ask whether a firm used AI. It will ask what the system produced, who checked it, and where the proof is. We built our operating system so that the answer to all three questions already exists, signed, on the firm's own hardware.”

What does a defensible AI workflow look like inside a law firm?

A defensible workflow is one in which verification is enforced by the system rather than remembered by busy people, and the evidence exists from the moment of the interaction. The referral shows how little a written policy is worth once an unchecked statement is on the record. A workflow that would survive scrutiny has five properties.

• A signed, tamper-evident record of every AI query and response, created at the moment of the interaction rather than reconstructed after a complaint arrives.
• Cross-checking of substantive claims before anyone relies on them, so a single model's fluent error cannot pass into a letter unchallenged.
• Human confirmation gates on the steps that matter, including anything destined for a court, a client or a counterparty.
• Privileged material kept inside the firm's own perimeter for the whole of the work, never routed through services the firm does not control.
• Evidence that can be verified offline, so the firm's account of events does not depend on a vendor's availability, retention policy or goodwill.

Measured against that list, a general-purpose consumer chatbot fails on every point. It holds its records, if any, on someone else's infrastructure, presents a single model's answer with unearned confidence, and places no gate between generation and reliance.

How does a sovereign intelligence operating system change the position?

It changes the position by making the evidence and the controls structural rather than optional. Mickai is a Sovereign Intelligence Operating System, a SIOS that runs entirely on the firm's own hardware, on premise and air-gapped where the sensitivity of the work demands it, so privileged client material never leaves the building while the work is done. Every action taken on the system is cryptographically signed into the Open Audit Record, which is post-quantum secure, tamper-evident and verifiable offline. When a question arrives about what the system produced on a given date, the answer is retrieved rather than reconstructed, and its integrity can be demonstrated without trusting us or anyone else.

Verification is built in rather than bolted on. Our cooperative multi-model consensus substrate requires specialist sovereign models to agree before any sensitive action runs, so one model's confident hallucination is challenged inside the system before a human ever sees it, let alone a judge. Sensitive actions are further gated by voice biometrics tied to a hardware-held root of trust, so the record shows an identified person approved each step. The architecture spans 87 studios on one operating system, ten production-ready at launch and 77 in development, protected by 104 filed UK patent applications across 2,340 claims, though the value to a firm lies in the evidence, not the filings.

What should firms do before the next referral?

Firms should assume the regulator's next question will be evidential and prepare accordingly, starting with an honest audit of where AI already touches client work without a record. The firms that come through this period well will be those that can show, for any AI-assisted output, exactly what was generated, what was checked and who took responsibility. That is not a restriction on using artificial intelligence. It is the condition under which a regulated profession gets to keep using it.

Frequently asked questions

What happened in the May 2026 High Court ruling on AI?

The UK High Court raised concerns about misleading, AI-influenced statements in letters dated 30 March and 14 April 2026, found a prima facie breach of the duty not to mislead the court, and referred the solicitor to the Solicitors Regulation Authority. The ruling treats unverified AI output as a conduct matter.

Does the ruling mean solicitors cannot use AI?

No. The duty engaged was the long-standing duty not to mislead the court. AI remains usable where outputs are verified before reliance and the firm can evidence that verification. The failure lay in the checking and the record, not in the decision to use the technology.

How does an audit trail protect a solicitor who uses AI?

An audit trail proves what the system was asked, what it returned, who reviewed it and when. Our Open Audit Record signs every action cryptographically with post-quantum algorithms, making the evidence tamper-evident and verifiable offline, turning a conduct inquiry from reconstruction into retrieval.

Why should legal AI run on the firm's own hardware?

Because privileged client material should never transit systems the firm does not control. Running the operating system on premise, air-gapped where required, keeps client data inside the firm's perimeter while every AI action is still recorded, cross-checked and gated before anyone relies on it.

What is MICKAI?

MICKAI is a Sovereign Intelligence Operating System, a SIOS that runs entirely on the customer's own hardware, on premise and air-gapped. Every action is cryptographically signed into the Open Audit Record, which is post-quantum secure, tamper-evident and verifiable offline. It brings together 87 studios on one operating system, with ten production-ready at launch and 77 in development, and it is protected by 104 filed UK patent applications across 2,340 claims, filed rather than granted.