What generative AI is doing to legal work and to the conduct of litigation: verification duties, AI-generated submissions, and the substantive law catching up.
In the same fortnight as OpenAI’s machine-generated proof of Navier–Stokes, the EAT told a litigant in person what a 300-page ChatGPT skeleton needs before it is filed: a reader who understands it. AI as a tool is not the problem here. The problem is a document handed up unchecked, and possibly not understood by the party filing it; the judgment lists what the EAT will do about one.
Falk and Tsoukalas's AI Layoff Trap shows that the only thing that stops over-automation is making each dismissal more expensive (the authors' “marginal instruments”). As currently drafted, Section 139(1)(b) ERA 1996 lets an employer replace an employee with a model and pay only a capped redundancy package; the most elegant fix open to a UK parliament is to redefine what counts as redundancy — turning the substitution from a safe exit into an exposed one.
The EAT in Edward holds that rule 62 reaches terms agreed in writing at any time, and survives s.144 of the Equality Act — but not a party who has told the tribunal, before the order is made, that he has changed his mind.
A lecturer dismissed for what he said to students never mentioned Article 10 before the tribunal. The EAT has held that speech alone does not make the Convention shout out, and in doing so has fixed the boundary between the two lines of authority on points a tribunal must take for itself.
The Bar Standards Board's May 2026 AI Guidance treats generative AI as a form of outsourcing under rC86, leaving Core Duty 7 unaffected. A practitioner reflection on Ayinde, Ndaryiyumvire and Munir, and the verification gap that separates law from mathematics.
A new Hart study from Sarah Fraser Butlin KC, Catherine Barnard and Maayan Menashe diagnoses what is wrong with the Employment Tribunal — and proposes a family-law-inflected reframe of the whole system, with a three-track adjudication structure, an Employment Resolution Service and an end to the formal grievance procedure as a precondition to litigation.
The Joint Presidential Guidance on panel composition implicitly relies on a mathematical theorem it never names. Condorcet's Jury Theorem explains why lay members improve outcomes — and the parallel with jury equity clarifies what Employment Tribunal panels are not.
The EAT in Tarbuc v Martello Piling Ltd reminds practitioners that Section 111A protection is not a procedural 'get out of jail free' card for employers who ambush employees with settlement offers.
UK indirect discrimination law demands that employers justify disparate impact. But when the decision was made by an opaque algorithm, the justification defence may be structurally unavailable.
A new economics paper models how AI reshapes firms from pyramids to diamonds. The implications for redundancy law, workforce planning, and the junior lawyers who may never be hired are worth taking seriously.
Garicano, Li & Wu's new research on task bundling provides a rigorous framework for understanding which legal jobs AI will displace. And which it won't. Employment law, with its deep interweaving of codifiable and contextual tasks, is a textbook strong bundle. The rumours of our death are greatly exaggerated.
The EAT has granted a restriction of proceedings order against a litigant who brought over 50 unsuccessful ET claims. A short guide to the three mechanisms, RPO, CPO, and CRO, available to restrain vexatious litigants in the tribunal system.
Harvard Business Review's latest research finds AI tools increase rather than reduce workload. For practitioners already navigating the Ayinde duty to verify AI output, this raises an uncomfortable question: if AI accelerates the pace of work, who has time to check it?
A cross-sectoral analysis comparing the flaws of "point estimates" in financial forecasting (OBR fiscal headroom) and litigation risk assessment (percentage prospects), arguing for the use of uncertainty bands instead of False precision.
Clients often ask how I arrive at a '60%' or '70%' chance of success. This article breaks down the methodology behind legal probability.
Practice: AI & Citation of Authorities · Algorithmic Management & AI · Redundancy