Can insolvency practitioners use AI on case data?

Yes. AI can accelerate claims review, records reconstruction and reporting across an insolvency caseload, provided three conditions hold: the estate's data stays on infrastructure the practice controls, each appointment's data is segregated, and every AI-assisted step writes to a sealed record that a court, a regulator or a challenging creditor can verify. Adjudication decisions and statutory reports remain the officeholder's own; AI drafts and checks.

The question matters in 2026 because insolvency work is document-heavy in a way few professions match, appointments do not schedule themselves around capacity, and the profession's duties make consumer AI services a poor fit for estate data. Practices that answer the question deliberately will move faster than those that let it be answered informally at desk level.

What data does an insolvency practitioner actually take on?

On appointment, an officeholder takes control of a company's entire information estate: books and records, employee data relevant to preferential claims and TUPE questions, creditor claims with their supporting documents, director conduct evidence gathered for misfeasance and disqualification reporting, and often live litigation files. The breadth is the point. A normal business holds its own data; an insolvency practitioner repeatedly inherits other people's, appointment after appointment, as an officeholder with statutory duties.

Why is confidentiality harder in insolvency than in an ordinary business?

Because the duties run in several directions at once, and some of them oppose each other. In a pre-pack, competing bidders must not see each other's positions. In litigation, the estate faces adverse parties who would value sight of the officeholder's working papers. Employee data, creditor data and director conduct material each carry their own sensitivities. And the officeholder is regulated: recognised professional bodies supervise the profession under Statements of Insolvency Practice, and process records are the raw material of that supervision. Placing estate data in a service the practice does not control sits badly with every one of those duties.

Where does AI genuinely help an officeholder?

In the document-heavy middle of every appointment, where the volume is real and the judgement is not yet engaged.

• Claims review: summarising creditor claims and checking them against the company's records, flagging discrepancies for the adjudicator.
• Records reconstruction: organising incomplete books, building chronologies and mapping transactions when records arrive in poor condition.
• Correspondence at scale: drafting routine creditor and employee correspondence for review before it is sent.
• Conduct reporting: assembling timelines and document references that feed the officeholder's assessment of director conduct.
• Report drafting: producing first drafts of progress reports and statutory documents for the officeholder to settle.

Each item accelerates the work. None of them is the decision.

Why must estate data stay inside the practice's boundary?

Because a public cloud AI service places estate data on infrastructure the officeholder does not control and cannot fully evidence. A practice cannot then say with confidence where a claim file went, who could access it or what was retained, and an officeholder's duties are not discharged by a vendor's assurance. On Mickai, a Sovereign Intelligence Operating System running offline on hardware the practice owns, a zero-egress perimeter means the estate's data has no outbound route at all. The confidentiality position becomes structural rather than contractual, which is a far easier position to defend.

How are appointments kept separate?

Deliberately, because conflicts are structural in insolvency: the same practice may hold appointments touching related parties, competing creditors or opposite sides of a transaction. This is why we treat segregation as an architectural requirement rather than a policy. Each appointment's documents, indexes and outputs are held apart, access is limited by role within the case team, and every access and every AI action is recorded against the specific appointment. The aim is not only to prevent cross-contamination but to be able to prove it never happened, because an allegation of leakage between matters is easier to make than to disprove without a record.

What will a court or a challenging creditor ask about an AI-assisted process?

Insolvency decisions are routinely challenged, and the officeholder must be able to evidence process: what was reviewed, on what basis a claim was admitted or rejected, and who decided. An AI-assisted process should strengthen that evidence, not blur it. Every action on Mickai is sealed to a post-quantum signed audit ledger bound to hardware-attested identity, and the record verifies offline, so the practice can show precisely what the system summarised, what it drafted and where the officeholder's own judgement was applied. A challenge then meets a complete record rather than a reconstruction from memory.

“An officeholder whose decisions are built to be challenged needs an AI record built to be examined.”

The sealed record and the offline architecture behind it are set out at /sovereign-ai, and the film at /film shows the interface at work.

Frequently asked questions

Can I paste creditor claims into ChatGPT to speed up adjudication?

Doing so places estate data on infrastructure the officeholder does not control, with retention and access the practice cannot fully evidence. Given the duties owed to creditors, the court and the regulator, the defensible route is AI running on infrastructure the practice owns, where the same acceleration is available without the exposure.

Who is responsible if AI mis-summarises a claim and it is wrongly admitted?

The officeholder. Adjudication is the officeholder's function and no drafting assistance changes that. This is why AI belongs in the checking and drafting layer, with a record showing the claim, the source documents and the human decision, so an error can be traced and corrected rather than buried inside an unlogged process.

Do I have to tell the court or creditors that AI was used?

There is no general rule requiring disclosure of drafting assistance, and practice in the profession is still forming. The safer position is readiness: an officeholder who can show from a sealed record exactly what AI did on a matter can answer any question a court, a regulator or a creditors' committee raises, whenever it is raised.

How is data kept separate between different appointments?

Through segregation built into the deployment. Each appointment's documents, indexes and outputs are held apart, access is limited to the case team, and every access and AI action is recorded against the specific matter. The record does double duty, preventing cross-use in practice and proving afterwards that it did not occur.

Can AI write statutory reports for an insolvency practitioner?

It can draft them. Progress reports, conduct assessments and statutory filings remain the officeholder's responsibility, and the judgements inside them must be the officeholder's own. AI usefully assembles the chronology, the figures and the first draft; the settled text, and everything it asserts, belongs to the officeholder who signs it.