A CaseFlow Automation Briefing

    THE WEEK IN AI

    Specialist AI workflows for claims and legal teams.

    Edition 9
    13 Sep to 20 Sep 2026
    Claims · Legal · Credit Hire
    This WeekFront Page · Theme · PERIMETER

    The bar for checking AI output has just reached people with no legal training at all.

    An Employment Appeal Tribunal judge rejected a 300-page AI-drafted appeal from a litigant in person and ruled that using AI is no excuse for an inaccurate or non-compliant document, whoever files it. The same week, insurers reported AI-manipulated evidence moving upstream into onboarding, and the FCA kept asking whether its own remit reaches the tools people use to reach a decision.

    Top Stories

    5 to know
    01

    A litigant in person filed a 300-page AI skeleton argument. The judge said the rules do not bend for that.

    Marcus Pilgerstorfer KC, sitting in the Employment Appeal Tribunal, ruled on the case of Terence Hancox, who filed a 132,000-word AI-drafted appeal document. The judge said personal responsibility for AI-drafted content, already established for solicitors, applies equally to litigants in person: check accuracy, check citations, keep only what is relevant. A repeat is likely to be rejected outright.

    CaseFlow angle. Extend your document intake checks to everything that reaches a live file, not only what your own team drafts. A litigant in person’s AI-drafted correspondence or disclosure needs the same verification pass as anything your firm produces internally.
    02

    Insurers say AI-faked evidence has moved out of claims and into onboarding.

    Insurance Post reported record levels of AI-manipulated document fraud, including forged repair invoices and number plate swaps on photographs of genuine accidents, now appearing at policy inception and mid-term adjustment as well as at the point of claim.

    CaseFlow angle. Provenance checks on repair invoices and damage photography can no longer sit only at the claims desk. Build the same verification step into intake and MTA workflows, because that is where the fraud is moving to.
    03

    The two leading AI models now cost the same on the label. The real bill still differs by four times.

    GPT-6 Astra and Claude Fable 5.1 launched within days of each other, both priced at 10 dollars in and 50 dollars out per million tokens. Independent benchmarks disagree with both vendors’ own claims about which model is actually ahead, and token efficiency, not sticker price, is what drives the real cost per task.

    CaseFlow angle. Do not choose a model on published pricing or a vendor’s own scoreboard. Benchmark your own document types, drafting or citation-checking tasks, and price the real token cost before switching or standardising.
    04

    Meta’s new AI agent takes a goal, not a prompt, and keeps working for days without you.

    Meta’s Muse product accepts standing objectives rather than single instructions and runs continuously inside a confidential virtual machine that Meta says even it cannot inspect. A separate monitoring layer flags payments, logins and sensitive data transfers for approval before they happen.

    CaseFlow angle. Any standing agent given access to a live claims or case file needs the same boundary, whichever vendor builds it: a defined scope, a named approver for anything sensitive, and a log that proves where it stopped.
    05

    The government’s own figures say a claim still takes 55.9 weeks to reach trial. That number is the funding cost on every disputed hire file.

    The Ministry of Justice’s Civil Justice Statistics for April to June 2026 put the median wait to trial for fast, intermediate and multi-track claims at 55.9 weeks, faster than a year ago but slightly slower than the previous quarter. Small claims sit at 41.0 weeks, longer on both comparisons.

    CaseFlow angle. Quantify what a year of hire funding costs on your own average disputed file and put that number next to whatever is actually being argued about. It is usually the stronger argument for resolving inside a framework than continuing to litigate.

    Claims & Legal Desk

    Claims · Legal · Credit Hire

    Sector signal for claims teams, law firms, and credit hire.

    A credit hire dispute scheme that was expected to go live this month still has not been confirmed as live by anyone.

    GTA ADR Phase 2, binding arbitration for disputed credit hire invoices under 10,000 pounds, was expected to open more widely in September 2026 after a successful 2025 pilot. As of 20 September, no source, including GTA and Verisk’s own channels, has confirmed a go-live date. The scheme remains voluntary for GTA members, and referral remains elective, case by case: it binds only once a case has actually been referred in, not automatically.

    CaseFlow angle. Do not tell clients or colleagues the scheme is live until you have checked the GTA or Verisk page directly. An expected reform that quietly slips is still a planning fact, and the silence is now itself the story worth watching.

    Motor claims inflation is reaccelerating on fresh official data.

    ONS confirmed CPI at 3.1 percent for the year to August, up from 2.9 percent in July. EY forecasts the motor sector’s 2026 net combined ratio at 108 percent, up from 102 percent in 2025. ABI data shows the average accidental damage claim reached 3,699 pounds in the first quarter of 2026, an 8 percent rise in a single quarter.

    CaseFlow angle. Feed the combined ratio and average claim figures into your own reserving conversation now, not at renewal, because both point the same direction.

    Ombudsman complaints about insurers rose by nearly a quarter in a year.

    Insurance complaints referred to the Financial Ombudsman Service rose 23 percent year on year in the second quarter of 2026, per analysis from Insurance DataLab.

    CaseFlow angle. Read this alongside your own complaint-to-claim ratio. A rising sector trend is not evidence your own handling has slipped, but it raises the bar for what “good” now needs to look like under Consumer Duty.
    Editor's Column

    The CaseFlow Take

    For ten weeks this column has argued that the person signing an AI-assisted document owns what is wrong with it. This week a tribunal took that principle somewhere we had not written it yet: onto someone with no legal training, no professional indemnity insurance, and no duty of care to anyone but themselves. The judge did not soften the standard for a litigant in person. He held it steady.

    We build the tool that sits underneath exactly this problem for credit hire and claimant firms, so we do not get to treat “AI output is not verified” as someone else’s compliance gap. Every authority CreditHire Assist puts in front of a handler carries a pinpoint to the paragraph relied on and a verified flag set by a person who opened the judgment. That is not a feature we added because it sounded safe. It is the only way the tool earns the right to be trusted with the argument at all.

    Untrained is not unaccountable.
    CaseFlow Automation

    CaseFlow Automation is a family of specialist software products for the claimant industry and regulated advisory firms.

    Contact

    Cyber Essentials certified

    Cyber Essentials certified. Read our security statement.

    © 2026 CaseFlow Automation Ltd. All rights reserved.

    ICO registered, UK Information Commissioner's OfficeICO Registration: ZC013423