The Dekagram: 16th November

News

16/11/2026

This week we welcome Hannah Ritichie and Charlie Blair as contributors to the Dekagram. In an example of the breadth of the work undertaken by the team, Hannah brings further news of the use of artificial intelligence in preparing for hearings, whilst Charlie considers the first UK case on notice provisions under the Montreal Convention.

Using and Misusing AI: Hancox v Sutherland [2026] EAT 139

The Employment Appeal Tribunal (“EAT”) commented on the use of generative artificial intelligence by litigants, particularly litigants in person, when preparing documents for submission to courts and tribunals.

In Hancox v Sutherland & Others [2026] EAT 139 Marcus Pilgerstorfer KC, sitting as a Deputy Judge of the High Court on 17 September 2026, set out guidance on a question that is coming up more and more often: what responsibilities fall on litigants, and particularly litigants in person (“LIPs”), who use generative AI to prepare documents for the Employment Appeal Tribunal? The same principles can be applied to civil cases in general.

Background

The Appellant previously worked for the National Farmers’ Union. He was dismissed in February 2020 and brought claims that included detriment, automatically unfair dismissal and disability discrimination. In 2021, EJ Lloyd struck out his claim against four individual respondents. The Judge found that he had conducted the proceedings in a scandalous, unreasonable and vexatious manner through a social media campaign on LinkedIn, and that a fair trial was no longer possible. The Appellant had made more than 145 LinkedIn posts about the subject matter of his claims.

The appeal was listed for an inter partes preliminary hearing on 29 July 2026. On the Monday before the hearing, the Appellant filed his “skeleton” argument. It ran to 300 pages and contained almost 132,000 words. The Court said that the word “skeleton” was “entirely inapposite” for it.

The covering email made things worse. The Appellant said he did not expect the Judge to read every page before the hearing. He also invited the Respondents to identify, in writing and by 4pm the next day, any factual inaccuracy, incorrect quotation, erroneous bundle reference or misrepresentation in the document. He added that a general complaint that the document was too long, or that there had not been enough time to read it, would not be good enough.

At the hearing the Appellant explained that he had used ChatGPT to create the document because he had to prepare it quickly. He accepted that much of it lacked credibility, and ultimately no reliance was placed on it.

The Legal Framework: No Lower Standard for Litigants in Person

The judgment begins by reminding litigants of the general position. The EAT Rules and the Practice Direction (“PD”) apply to every litigant, represented or not. The Judge relied on the well-known case of Barton v Wright Hassall LLP [2018] 1 WLR 1119. This case held that being unrepresented may justify allowances in case management and in the conduct of hearings, but “will not usually justify applying to litigants in person a lower standard of compliance with rules or orders of the court“. Those principles apply equally in the EAT. It is generally reasonable to expect LIPs to familiarise themselves with the EAT Rules and the PD, which are available online and drafted to be comprehensible to non-lawyers. 

All litigants before the EAT, whether represented or not, must comply with the EAT Rules and PD. The requirements for skeleton arguments in the EAT, set out in PD para 11.6.6, are reproduced in full. A skeleton should generally be between 5 and 15 pages long. The EAT, like the Court of Appeal, expects these requirements to be “rigorously observed”. Non-compliant skeletons may be rejected, and other sanctions are available, including those under PD para 11.6.7 and Rule 26 EAT Rules. As described above, the Appellant’s “skeleton”, to say the least, fell short.

The AI Analysis

The Judge accepted that AI brings opportunities as well as risks and recognised its potential to assist those who have no access to professional legal advice. Notwithstanding this, the Court listed the risks that AI-generated material may be:

  • unfocused, excessive or not directly relevant;
  • non-compliant with procedural requirements;
  • the product of poor input material or poor prompting;
  • misleading as to the factual, evidential or legal position;
  • raising new points on appeal that were not argued at first instance;
  • affected by in-built bias; or
  • containing hallucinations, such as made-up case citations.

The judgment then reviews the authorities, beginning with R (Ayinde) v London  Borough of Haringey [2025] 1 WLR 5147, which was mainly concerned with regulated professionals. In Ayinde, the Court observed that freely available tools such as ChatGPT “are not capable of conducting reliable legal research“. Those using AI must therefore check its output against authoritative sources, such as legislation.gov.uk, the National Archives judgment database and the official Law Reports. The Judge held that the same risks to the administration of justice arise when a litigant in person uses generative AI.

In Taiwo v Homelets of Bath Ltd [2025] EWHC 3173, Constable J observed that reliance on false citations “is just as unsatisfactory when presented to the Court by a litigant in person“, although the sanction may differ. The Judge also referred to Rafique v Revenue and Customs Commissioners [2026] UKFTT 673 (TC), where the First-tier Tribunal held that all parties, whether legally trained or not, must ensure that what they submit is factually correct. The Court also relied on the Artificial Intelligence – Guidance for Judicial Office Holders (31 October 2025), which says it is appropriate to ask a litigant what accuracy checks have been carried out, as well as:

  • Baker LJ’s statement in Re D (A Child) (Recusal) [2025] EWCA Civ 1570 that “all parties – represented and unrepresented – owe a duty to the court to ensure that cases cited in legal argument are genuine“; and
  • the Presidential Guidance – Applications for interim relief (22 June 2026), which notes that AI-generated submissions are often “too long and complex” and fail to focus on the key points.

The Minimum Standard

Paragraph 32 is likely to be the passage that is most helpful to practitioners, and one could expect reference to [32] to be frequent. LIPs who use AI should ensure, at a minimum, that every document they submit:

  1. complies with the applicable procedural rules, such as, in the EAT, PD para 3.8 for grounds of appeal and PD para 11.6 for skeleton arguments;
  2. has been checked for accuracy as thoroughly as the litigant or representative is reasonably able, so that factual, evidential and legal points (including references to authorities) are correct and the tribunal is not misled; and
  3. contains only relevant points, focuses on the central or best arguments, is easy to understand and avoids undue repetition.

The judgment adds two further points:

  1. Simply submitting the product of generative AI to a court or tribunal, or placing the onus to check a document on an opponent, is not acceptable, and judicial enquiries and possible sanctions are likely to follow.
  2. Given the EAT will not generally consider points that were not argued below, AI-generated documents must not slip new points in without a proper application under PD para 8.13.1.

As to sanctions, the Court noted that Ayinde identified a range of responses. Some of them, such as referral to a regulator or a wasted costs order, depend on professional status. Others do not, such as strike-out, costs sanctions, public admonishment and, in the most egregious cases, contempt proceedings or a police referral for perverting the course of justice. The EAT has similar powers. Where the litigant in person is in fact a regulated professional, regulatory obligations may still apply.

The Judge also warned that using AI may lead a tribunal to be cautious about how much weight it gives to a document or to evidence, and cited Godwin v Godwin [2026] EWHC 923 (Ch), where ChatGPT had been used to prepare witness statements.

Application

Inevitably, the Court was satisfied that the Appellant had not carried out the necessary checks. This was clear from the covering email and from the Appellant’s own assessment of the document at the hearing. The attempt to shift the job of checking onto the Respondents, particularly for inaccuracies, and on an unreasonable timetable, was “another unacceptable feature”.

Comment & Conclusion

Hancox does not ban the use of AI; indeed, the judgment expressly states that its use is “not generally prohibited for tasks such as preparing skeleton arguments”. It does, however, serve as a warning, extending the principle in Ayinde beyond the professions. Whoever puts a document before a tribunal is responsible for it, however it was produced.

For practitioners, the case is a useful authority when facing a LIP who serves voluminous, AI-generated material and then asks the represented party to correct it. Paragraphs 33 and 86 make clear that this burden does not move to the other side.

About the Author

Hannah Ritchie was called in 2022 and specialises in civil and commercial litigation, with particular expertise in civil fraud and personal injury. As part of her personal injury practice, Hannah is regularly instructed on employers’ liability, public liability, credit hire and trials concerning allegations of exaggeration of injuries, fundamental dishonesty, and deliberate layering. In her spare time, she enjoys surfing and skiing.

Timely notice of complaints under the Montreal Convention

Personal injury practitioners will be well acquainted with Article 17 of the Montreal Convention and the two-year limitation period under Article 35. The Montreal Convention, however, also covers claims for the destruction, loss of, damage to and delay of baggage and cargo. All of these share the same two-year limitation period under Article 35 but claims for damage and delay are also subject to the notice of complaint requirements in Article 31. 

Under Article 31, receipt of checked baggage or cargo without complaint is prima facie evidence that they have been delivered in good condition. The deadline for complaints under Article 31(2) comprise:

  • Checked baggage damage: written complaint forthwith after discovery and at the latest within 7 days of receipt. 
  • Cargo damage: written complaint within 14 days of receipt.
  • Baggage or Cargo delay: written complaint within 21 days of being placed at the Claimant’s disposal. 

Complaints must be in writing (Article 31(3)). If no complaint is made within the above timeframes, Article 31(4) provides that “no action shall lie against the carrier, save in the case of fraud on its part”.

What constitutes a complaint under Article 31? 

This may appear to be a simple question but it has produced a surprising amount of litigation in other jurisdictions. The recent decision in Chemidex Pharma Ltd v United Airlines Inc [2026] 11 WLUK 40 is a cautionary tale and appears to be the first reported English decision on what amounts to a complaint under Article 31 of the Montreal Convention. 

Brief background

United agreed to fly 26 pallets of prednisolone 20mg foam, a temperature-controlled pharmaceutical, from Houston to London Heathrow. The product had to be kept between 15°C and 25°C throughout storage and transit.

Once the shipment had arrived, the claimant’s agent, acting as consignee, wrote to United by email. The email noted temperature readings above 25°C at both Houston and Heathrow, it said “wondering if you can assist in investigating shipment”, “we understand the gap in temperature but just to clarify to the customer can you help”, and posed three questions. Subsequent testing showed that the product had been damaged by heat and could not safely be used.

United applied to strike out the claim on the basis that this email was not a complaint for the purposes of Article 31. The judge agreed, and the claim was struck out.

The email did not comply with Article 31(2) for the following reasons: 

  • its tone and language were investigatory, not of a complaint;
  • it described conditions in transit (namely temperature readings) but nowhere said that the goods had been damaged or that damage was being claimed;
  • its opening sentence was a question and its second a statement of fact, and the remainder were describing the conditions in transit; and
  • read as a whole, the email conveyed no different meaning.

In reaching that conclusion, the judge drew together the following principles on the construction of Article 31:

  • The Convention should be interpreted uniformly across signatory states.
  • The test is objective and practical. 
  • Describing the condition of the goods is not enough; there must be a reference to damage. “5 boxes wet” (Oriental Insurance Co Ltd v BAX Global Inc), “beans very hot” (Yoly Farmers Corp v Delta Air Lines Inc) and “freight is bent/broken” (OEC Group (NY) Inc v China Cargo Airlines Ltd) were each held insufficient. 
  • The complaint must cover the damage on which the claim is based (Western Digital Corp v British Airways Plc [2001] QB 733).
  • Its purpose is to enable the carrier to make enquiries (Fothergill v Monarch Airlines Ltd [1981] AC 251).
  • It need not expressly assert that the carrier is liable. 
  • A tentative suggestion that the cargo may be unusable is insufficient. 

Comment

Chemidex confirms that a complaint is a matter of substance as well as form. The CJEU case of Finnair Oyj v Keskinainen Vakuutusyhtio Fennia (C-258/16) EU:C:2018:252, [2018] 1 W.L.R.

4407 takes a broad view of form (a complaint recorded in the carrier’s computer system can be “in writing”) but makes clear that the purpose of a complaint is to inform the carrier that damage has occurred. The Chemidex email did not do so.

In practice, a complaint should:

  • be in writing (email will do) and sent within the Article 31(2) time limits; and
  • state that the goods are damaged, rather than merely reporting conditions such as temperatures. 

About the Author

Charlie Blair was called in 2025 and successfully completed pupillage at Deka Chambers in 2026. Prior to pupillage, he worked as a paralegal at Allen & Overy and at Stewarts. In the latter role, he represented numerous disabled clients before the Social Security Tribunal. He accepts instructions across Chambers’ main areas of practice and has a strong commitment to pro bono work.

Featured Counsel

Hannah Ritchie

Call 2022

Charlie Blair

Call 2025

Latest News & Events

The Dekagram: 16th November

This week we welcome Hannah Ritichie and Charlie Blair as contributors to the Dekagram. In an example of the breadth of the work undertaken by the team, Hannah brings further news of the use of artificial intelligence in preparing for hearings, whilst Charlie considers the…

Mediation vs Litigation: a lunchtime guide for litigators

Join Deka Chambers barrister and mediator Laura Elfield and Tessa Herman, an experienced personal injury solicitor and mediator for a whistle stop tour on mediation. Laura and Tessa, both NHS Panel Mediators, will look at when and why to opt for mediation, explore the different forms of mediation, give tips on getting…

Deka Chambers scoops up two awards at the Chambers UK Bar Awards 2026

We are thrilled that Andrew Warnock KC has been awarded Personal Injury Silk of the Year and Deka Chambers Personal Injury Set of the Year at the Chambers UK Bar Awards at a ceremony on Thursday evening 12 November.  In the latest Chambers UK Bar…

© Deka Chambers 2026

Search

Portfolio Builder

Select the expertise that you would like to download or add to the portfolio

Download    Add to portfolio   
Portfolio
Title Type CV Email

Remove All

Download


Click here to share this shortlist.
(It will expire after 30 days.)