TL;DR:

• London competition associate Larissa Meredith-Flister has built an AI tool — Opposing Counsel Review — that stress-tests legal arguments by deliberately attacking the reasoning, exposing evidential gaps and modelling how a sceptical judge would respond.
• The tool is free via Lawvable and works as a skill on Claude, ChatGPT and Gemini; it is a personal project outside her day job at litigation boutique Charles Lyndon.
• The build is a pointed counter-example to the dominant “AI as summariser” narrative in legal tech — here AI functions explicitly as an adversary, closer to red-teaming than to document automation.

Meredith-Flister’s framing is sharp: “It’s not a proofreader or a summariser — it’s an adversary, and that’s the point.” The tool generates qualifying questions tailored to the input document (witness statement, skeleton argument, structured reasoning), then dismantles the argument once answered.

A UK practitioner-led pattern

The notable pattern here is UK lawyers building their own AI tools in their spare time and releasing them into shared skill catalogues rather than going through the traditional legal-tech vendor route. Tris Sherliker of Bird & Bird launched BunTool for court bundles earlier this month; Meredith-Flister is releasing Opposing Counsel free. Networks like Legal Quants and RepresentAI — which connect lawyers with coding skills — appear to be seeding a micro-tooling movement inside UK practice.

That matters for three reasons. First, it puts adversarial testing into routine case preparation rather than leaving it as expensive consultancy. Second, it demonstrates that high-stakes legal workflows can be instrumented with current-generation LLMs when scoped carefully. Third, it pressures incumbent legal-tech vendors, whose flagship products are still largely drafting, summarisation and contract-review tools.

Competition-law worked example

Meredith-Flister illustrated Opposing Counsel’s output with a competition-law submission that argued a dominant firm’s exclusive supply agreements amounted to abuse of dominance on a 60 per cent market share. The tool exposed that the submission had not defined the relevant market, had conflated exclusive agreements with anticompetitive effect, and had offered no economic evidence of foreclosure — the sort of structural gaps a UK competition tribunal would focus on first. The point, she said, is that “the argument may read persuasively” even when the underlying analytical work is missing, and peer review from colleagues tends to be polite rather than adversarial.

Looking forward

Solicitors in litigation-heavy practices are likely to adopt tools of this kind quickly, particularly for high-consequence arguments where the cost of a missed structural flaw is a lost case. The question for UK bar regulators is whether using a free AI adversary introduces professional-conduct or confidentiality risks when the underlying LLMs sit outside firm infrastructure — an unresolved issue that the Solicitors Regulation Authority has yet to rule on.