The file and the office

AI's impact on the licensed insolvency and restructuring profession worldwide, covering the United Kingdom, the United States, the European Union, Australia, Singapore and India, as at October 2026.

Insolvency practitioners have a better answer to artificial intelligence than most professionals. Only a licensed practitioner can take an appointment as liquidator, administrator or trustee, carry the statutory duties that come with it and answer to a court. A model holds no licence and cannot be appointed, so the profession's core is protected by law rather than by the limits of the technology.

That protection is where the danger sits. Because the model cannot hold the office, everything it produces becomes the practitioner's the moment it is signed. On 3 June 2026, in Lnu v. Blanche, the US Court of Appeals for the Ninth Circuit put the principle plainly: the rules are not violated at the point of research and drafting, but at the point of signing and filing.

That court sanctioned two attorneys whose briefs cited cases that did not exist, and suspended them from practising before it for six months. The AI Hallucination Cases database, maintained by the researcher Damien Charlotin, now lists more than 2,100 court decisions worldwide involving AI-fabricated material.

The licence keeps the model out of the office. It does not keep the model's errors out of the file, and the file carries the practitioner's signature. Meanwhile the routine casework that trained practitioners to catch those errors is the work AI is absorbing first. AI will increasingly own the file. The office stays human, and so does the blame.

Six risks, ranked by how badly they bite

Worst first; the evidence follows.

1. Liability at the signature. An AI error becomes the office-holder's error the moment the document is signed or filed. Sanctions now reach the right to practise, not just costs: the Ninth Circuit suspended two attorneys from practising before it for six months over AI-fabricated citations in June 2026.

2. The vanishing training ground. Document intake, claims reconciliation and first drafts are the work AI takes first, and they are where trainees learned to spot a suspect transaction. The profession has no settled answer to how the next generation of licensed practitioners will be trained.

3. The fee argument. Creditors, committees or the court approve an office-holder's pay. When automation cuts the hours a task takes, a matter that took 400 hours invites the question of why it did not take 250.

4. Confident fabrication. In peer-reviewed testing by Stanford's RegLab, the leading AI legal-research tools from LexisNexis and Thomson Reuters each produced false answers between 17 and 33 per cent of the time, and a fabricated authority looks identical to a real one.

5. Platform dependence. Case management is concentrated in a few specialist platforms. Live case files, statutory deadlines and creditor ledgers are hard to move mid-appointment, which hands suppliers pricing power and leaves practitioners exposed if a supplier changes its model or fails.

6. Regulation, slower than it looks. No jurisdiction has an AI rulebook for insolvency. The European Union's AI Act will impose its stricter rules on stand-alone high-risk systems from 2 December 2027, but the binding rules everywhere remain the existing duties of the licensed practitioner.

Wide but shallow: AI adoption in insolvency practice

The clearest measure of adoption comes from the United Kingdom. On 2 September 2026 R3, the UK trade body for restructuring and insolvency professionals, published with the technology firm Alph4 what it describes as the first comprehensive review of how the profession uses AI and automation. Of respondents, 81 per cent had digitised document intake and 52 per cent used tools such as Microsoft Copilot or ChatGPT, but fewer than one in ten used machine learning or AI agents.

R3 drew its own conclusion: regulators should provide clearer guidance on responsible AI use, including expectations on data security, human oversight, quality assurance and disclosure.

The work that is moving is administrative and document-heavy: intake, claims reconciliation, creditor correspondence and statutory reporting. Most of it arrives as features inside existing case-management platforms rather than as standalone AI products, and those platforms are few. Aryza says its Insolv software already manages more than 60 per cent of monthly appointments in UK personal insolvency. A practice that runs its statutory diary, creditor ledgers and compliance packs on one platform cannot easily leave it in the middle of an appointment.

Note. Almost everyone has digitised the paperwork; almost nobody has put reasoning tools into live casework. The gap is held open by uncertainty over liability, not by doubt about the technology.

Who signs carries the risk

The record of AI failure in legal work has moved from embarrassment to loss of livelihood. In Lnu v. Blanche the Ninth Circuit imposed monetary sanctions and a six-month suspension from practice before the court, explaining that the gravity of the discipline followed from the attorneys' repeated failure of candour about where the fabrications came from. In Australia, the Victorian Legal Services Board varied the practising certificate of a solicitor who had tendered AI-generated citations to the Federal Circuit and Family Court, so that he could no longer practise as a principal lawyer; the Board called its action an Australian first.

Courts are acting in matrimonial and commercial cases alike. In Prososki v. Regan, the Nebraska Supreme Court struck a brief full of nonexistent cases and fictitious quotations, dismissed the appeal, and referred the lawyer who signed it to the state's Counsel for Discipline.

Insolvency sharpens every part of this. The office-holder owes duties to a body of creditors who did not choose the practitioner and cannot see the tools being used, so a mis-adjudicated claim moves money between real creditors. Much of the output is filed on oath or certified to a court. The distinction courts draw is between a tool that was used and a judgement that was exercised, and only the second protects the person who signed.

Flight 447 and the claims ledger

On 1 June 2009 Air France Flight 447 was lost over the Atlantic with 228 people aboard. According to the French air accident investigator, the BEA, ice obstructed the speed probes, the speed readings became erroneous and the automatic systems disconnected. The crew did not identify the approach to the stall, failed to diagnose the stall itself, and made none of the inputs that would have made recovery possible. Automation had not removed the pilots' responsibility. It had removed their practice, and then tested them at the worst moment.

The insolvency version is quieter but built the same way. A trainee who never reconciles creditor claims by hand or builds a statement of affairs from the ledgers cannot tell when an automated reconciliation is wrong. A mid-career practitioner becomes an editor of machine drafts rather than their author. A senior practitioner keeps the judgement personally but supervises staff who never acquired it, so the checking falls on fewer people just as estates grow more complex.

The warning signs inside a practice are visible: figures and citations in the file that nobody can trace to a primary record, turnaround times falling faster than checking, and juniors who cannot do the task when the tool is taken away.

Air France 447's failure modes, in an insolvency practice

Failure mode	On Air France 447, 2009	In an insolvency practice now
Loss of reliable inputs	Iced probes made the speed readings erroneous and the automatic systems disconnected	A tool changes model version, or reads bad ledgers, and its output stops being reliable without warning
Failure to recognise	The crew did not identify the approach to stall	A fabricated authority or a wrong reconciliation looks identical to a correct one
Failure to diagnose	The crew failed to diagnose the stall situation	A distress or valuation output that matches a first hunch stops being tested against the ledgers
Responsibility	No inputs were made that would have made recovery possible	The rule is breached at signing and filing, so the signature carries the liability

Note. Each failure mode pairs what happened in the cockpit, which is history, with the same failure on an insolvency file, which is a forecast.

The 400-hour question: fees under pressure

Insolvency is unusual in that the people paying are often not the people choosing. An office-holder's remuneration is scrutinised and approved by creditors, a committee or the court, and time costs must be justified. That makes the profession acutely exposed to a technology whose main effect is to shorten tasks. Each hour automation removes from claims, review and investigation is an hour a well-informed creditor can decline to pay for.

The value is migrating upward. What holds its price is the work only a licensed person can carry: judging whether a business is worth saving, negotiating with hostile stakeholders, defending realisations before a court and signing with personal liability attached. Volume personal insolvency, built on standardised, rules-based administration, faces the sharpest squeeze.

Governments are standardising the low end too. Singapore's Ministry of Law launched a revamped Simplified Insolvency Programme on 29 January 2026, administered by licensed insolvency practitioners. A standardised statutory process is the natural home of automation.

Insolvency careers: the licence and the thinning base

For now, hiring follows the caseload more than the technology, and caseloads in several markets are large. India's Insolvency and Bankruptcy Board reports that 69 corporate insolvency processes yielded resolution plans in April to June 2026 alone, and 1,484 since the Insolvency and Bankruptcy Code took effect. Rising case numbers can absorb a great deal of automation.

The licence remains the gate. In the UK, the Joint Insolvency Examination Board's own statistics show pass rates between 26 and 50 per cent across its corporate and personal insolvency papers in 2023 and 2024. That scarcity is what protects senior pay; the analyst and administrator roles beneath it have no such barrier.

The likeliest five-year effect is a decoupling of caseload from headcount at the base: firms clearing more cases per administrator, and junior intake growing more slowly than the work. New work exists at the edges, including the restructuring of AI companies themselves, where what an insolvent debtor actually owns in its models and data is still contested.

AI disruption risk by seniority in insolvency and restructuring

Level	Risk	Why
Junior: administrator, analyst, trainee	High	Days built from intake, statutory forms, claims logging and first-pass investigation, the work tools automate first
Mid: case manager, senior analyst	Moderate to high	Squeezed from below as junior work shrinks and from above as drafting and review are augmented
Senior: director, licensed practitioner	Low to moderate	Appointments, court-facing decisions and negotiation stay human; the exposure is liability for unchecked output
Partner	Low	Origination, reputation and sign-off risk; AI adds governance duties

Note. The protection is legal, not technical, and it sits at the top. The base of the pyramid has no statutory shelter at all.

Six jurisdictions, one answer

No jurisdiction has written a bespoke rulebook for AI in insolvency, and the postures differ. The United Kingdom works through existing licence conditions while reviewing its Insolvency Rules; the government's second review, published on 14 July 2026, asked how the rules should respond to developments in artificial intelligence, digital assets and cryptocurrencies, and closed on 6 October 2026. The European Union's AI Act leaves insolvency law untouched but regulates the tools, and the European Commission has confirmed that its rules for stand-alone high-risk systems apply from 2 December 2027. The United States governs by sanction, case by case.

Australia relies on existing professional duties, alongside a September 2024 government proposals paper on mandatory guardrails for AI in high-risk settings. Singapore has made a simplified, practitioner-run procedure part of the regime. India has no AI-specific insolvency rule, but its tribunals clear a volume of cases that makes automation attractive.

By different routes, every one of them reaches the same place: AI may inform a licensed practitioner's decision, but cannot be one.

How AI in insolvency is governed, by jurisdiction

Jurisdiction	Approach	The instrument that matters
United Kingdom	Existing licence and statutory duties; the profession has asked regulators for guidance	Second review of the Insolvency Rules, covering AI; consultation closed 6 October 2026
European Union	Horizontal AI law; insolvency law itself unchanged	AI Act: rules for stand-alone high-risk systems apply from 2 December 2027
United States	Judge-made, through sanctions at signing and filing	Court rules and decisions, such as Lnu v. Blanche (Ninth Circuit, June 2026)
Australia	Existing professional duties; mandatory guardrails proposed	Proposals paper on mandatory guardrails for AI in high-risk settings, September 2024
Singapore	Simplified process administered by licensed practitioners	Revamped Simplified Insolvency Programme, from 29 January 2026
India	No AI-specific insolvency rule identified	Insolvency and Bankruptcy Code: 69 processes yielded resolution plans in April to June 2026

Note. No jurisdiction has written an AI rulebook for insolvency. Every one of them, by different routes, leaves the answer with the licensed person.

What it comes down to

The fair rating for licensed insolvency and restructuring work is moderate disruption, rising to high for junior and routine analytical tasks, and held well short of transformation by a licensing monopoly no model can cross. On a scale where 100 means wholesale displacement, the profession scores about 45: above general professional services, well below fields whose core deliverable a model can produce. The base case, weighted at 55 per cent, is steady absorption between 2027 and 2032, with the junior and paraprofessional tiers thinning and the senior tiers holding.

The distinction that matters is between the file and the office. AI will increasingly own the file: the intake, the reconciliation, the first drafts and the routine correspondence that once filled a junior's week. It will not own the office, because someone must exercise independent judgement, owe duties to creditors and the court, and be accountable when things go wrong.

That leaves each practitioner a choice rather than a forecast. Using AI to do more judgement makes a practitioner more valuable as the tools improve; using it to avoid judgement, signing outputs that can no longer be independently checked, makes the same practitioner less valuable and more dangerous. In your own work, which documents carry your name that you could no longer rebuild by hand?

Drawn from TheRiskAgent's report AI Impact on the Insolvency / Restructuring (Licensed) Profession (September 2026); every figure here was checked against its original publisher. Produced with AI research tools and reviewed before release; reference material, not advice. The full report carries the detailed analysis.