On 8 September 2026 OpenAI published a 165-page proof that smooth solutions to the Navier–Stokes equations can break down in finite time. Up to 10,000 agents ran for 88 hours to produce it; it came with a formalisation in Lean, so that a machine could check every step. Ten days on, the Clay Mathematics Institute has not pronounced, a priority dispute is running; the mathematicians reading it are on the one question Lean cannot answer: whether the theorem proved is the problem posed.
Nine days after the proof, the EAT handed down Hancox v Sutherland [2026] EAT 139.
A litigant in person had filed a skeleton argument of 300 pages and almost 132,000 words, generated with ChatGPT, and invited the respondents to identify any inaccuracies in writing by 4pm the next day. Marcus Pilgerstorfer KC, sitting as a deputy judge of the High Court, did not treat the tool as the problem. The judgment opens by recognising what generative AI offers a party without access to legal advice, and says at [31] that there is no principled reason why a litigant in person should not use it. What it will not accept is the document arriving unchecked, or the checking being passed to the other side. The minimum at [32] has three parts: the document complies with the Practice Direction, it has been checked for accuracy as thoroughly as the litigant is reasonably able; it confines itself to the best points. At [33], submitting the product of generative AI without those checks, or placing the onus on an opponent, is not acceptable: enquiries and sanctions are likely to follow.
That is the Navier–Stokes position translated into procedure. A proof nobody in the field has read is a candidate, however many agents produced it; most candidates fail before one holds. The expert reading is what makes the work usable. A 132,000-word skeleton is a candidate too: the request that the respondents find its errors by 4pm the next day was an attempt to move the checking to the one place it cannot go.
The judgment also sets out, in one place, what the EAT will do when a document arrives in that state.
| Power | Source | Effect |
|---|---|---|
| Reject the skeleton | PD 11.6.6(a); [21] | A skeleton that is too long is refused and a compliant version required |
| Sanction non-compliance | PD 11.6.7; rule 26; [21] | Postponement, dismissal of the appeal, or costs |
| Cut the oral hearing | [21] | Less time for submissions already fully in writing |
| Applications on the Annex 2 form only | PD 7.3.1; [162] | Anything made by correspondence is not considered |
| Certify as totally without merit | London Underground Ltd v Mighton [2020] EWHC 3099 (QB); [146]–[148] | A hopeless application is recorded as such, the first step towards a restraint order |
| Limit correspondence | J v K [2019] EWCA Civ 5; PD 1.9.2; [165]–[167] | Email contact restricted if excessive correspondence continues, with the party’s health weighed first |
The world the judgment describes is the world the proof describes. The machines will produce more, faster, and some of it will be right. The rules of the tribunal, like the norms of the mathematical community, do not ask who produced the document; they ask who understood it before it was filed. For now that has to be a person who could have written a shorter one.
In practice
- Against a machine-generated skeleton, apply under PD 11.6.6(a) before the hearing for a compliant version and a page limit, citing Hancox at [32] and [33]. Decline any invitation to find the errors yourself; the judgment says the onus cannot be moved.
- Where correspondence is already heavy, ask for the Annex 2 order at the first case management opportunity. It is the order that stops applications arriving in pieces.
- For your own use of the tools, the standard is stricter: R (Ayinde) v London Borough of Haringey [2025] EWHC 1383 (Admin) and the BSB’s guidance under rC86. The Hancox minimum is the floor for every litigant, not the ceiling for counsel.
Table of Authorities
| Case | Citation | Proposition |
|---|---|---|
| Hancox v Sutherland KB → | [2026] EAT 139 | A litigant in person using generative AI must ensure the document complies with the Practice Direction, has been checked for accuracy as thoroughly as they reasonably can, and confines itself to the best points; the onus cannot be placed on the opponent. |
| J v K KB → | [2019] EWCA Civ 5; [2019] ICR 815 | A party’s mental ill-health is always a relevant consideration in the exercise of a procedural discretion, though not necessarily decisive; Underhill LJ’s general guidance at [39]. |
| London Underground Ltd v Mighton KB → | [2020] EWHC 3099 (QB) | Tribunals should consider and record whether an application is totally without merit; the certification is the foundation for a restraint order. |
| R (Ayinde) v London Borough of Haringey KB → | [2025] EWHC 1383 (Admin) | Regulated lawyers who file material generated by AI without verifying it breach their duties to the court; the Divisional Court’s range of responses runs from admonishment to referral and contempt. |
Alex acts in disputes like this one for claimants and respondents, instructed by solicitors or by direct access. How to instruct him or email the clerks.