Regulatory and market watch
Developments in legal AI, selected for their relevance to litigation teams, alongside our own analysis of what they mean in practice.
A disciplinary appeal drafted, attacked, verified and filed in three days. Then the complete session transcripts were parsed line by line, to establish what happened.
And the idea behind it may matter more to a twelve-lawyer litigation boutique than to the firm that just committed half a billion dollars.
Using construction disputes as the test case, the authors argue that AI retrieving the right records quickly is not yet an investigation a legal team can defend. What makes it defensible is a structured evidence base, a traceable process and professional review that leaves a record.
Sullivan & Cromwell, Cooley and Ropes & Gray are showing clients their own AI tools, and clients are asking where the saved hours appear on the bill. Firms that bill by the hour are squeezed from both sides: working faster cuts hours, and rates cannot rise fast enough to cover the gap.
All but one of the agent harnesses tested let the AI agent delete its own execution trace when asked, without tripping a monitoring guardrail, and tampering also emerged unprompted when models were chasing a better reward. For audits and incident reviews that depend on those logs, the authors advise recording them through a mechanism the agent cannot reach.
Alma, an immigration law firm built around AI and co-founded by a former Cooley associate, raised a $17M Series A and plans to hire experienced lawyers with Big Law backgrounds. Capital is going to firms that deliver legal work with AI as well as to the vendors selling tools to existing firms.
Noxtua raised about €100M in a Series C that makes German legal publisher C.H.Beck its majority shareholder, with Austria's MANZ also investing and early backers CMS and Dentons selling their shares while staying on as clients. Europe's legal publishers are buying the AI layer that sits over their own content.
Litera put Foundation, the firm-intelligence platform used by more than 300 law firms, inside ChatGPT Enterprise as a plugin, so lawyers can query their firm's experience and relationship records in plain language. Answers are limited to the records each user is permitted to see, which makes permissions and source records the part a firm has to get right.
OpenAI released Astra for Law, a legal configuration of GPT-6 Astra paired with an index of US case law, statutes and regulations, which Harvey and Legora can build on and which connects to Relativity, Clio, Intapp and Thomson Reuters. With the model vendor now supplying the research layer, specialist legal AI products have to show what they add on top of it.
New York courts are starting to decide whether AI prompts and model outputs are protected work product or discoverable, with matrimonial cases at the sharp end because forensic accountants now use AI in business valuations. Experts, and the lawyers instructing them, should assume the prompt history may be asked for.
Apple's trade-secrets suit against OpenAI escalated when Apple told the court that OpenAI was destroying evidence, while OpenAI asked the judge to deny a preliminary injunction. If the case reaches discovery, both companies may have to show how they document and protect confidential information.
Satrix ran a forensic audit of the complete session transcripts behind an AI-assisted legal appeal: 46 production agents, 192 citations checked to the page, 18 defects caught before filing, and a catalog of 12 failure and recovery pairs. The process audit and the full case study are published alongside.
Aavalynx raised €1.75M pre-seed for a dispute-intelligence platform aimed at enterprises, law firms and litigation funders, covering early case assessment, chronology building and metadata extraction, with every statement cited back to its source. Founded by a disputes lawyer out of Freshfields, Latham and Quinn Emanuel, it treats litigation exposure as a risk category to be priced.
Align Research retrieves real court opinions and marks the relevant passages without generating any text of its own, at $100 per task billable to the matter. Refusing to generate is the product claim, and it puts source reliability ahead of fluency.
Legora is acquiring Wexler, whose technology extracts and verifies facts from unstructured documents into source-traceable case records, and will run it as the fact layer beneath its agentic workflows. Fresh from a $600M round at a $5.6B valuation, the large platforms are buying evidence provenance rather than building another research surface.
Epiq is turning natural-language requests into actions inside eDiscovery workspaces, covering processing, search, review and production sets, with Claude behind its own reasoning layer and rollout from August across 2,600+ client organisations. The governance it ships with — permissions, workflow approvals, audit logs, action traceability — is as much the product as the model.
Nearly a quarter of law firm professionals would decline an offer from a firm without professional-grade AI tools, and roughly a third of corporate legal buyers are reconsidering outside counsel who cannot show AI-enabled value inside 12 months. Only half say their firm's AI strategy is visible in daily operations, which is the gap shadow AI grows in.
Federal civil pro se plaintiff rates rose from 11.33% to 16.94% after generative AI became widely available, and AI-drafted complaints run citation-dense yet get dismissed earlier, per O'Toole Scrivo. Defence costs move before the merits do, which puts a price on detecting AI-assisted filings early.
DocJuris moved past contract review with Workforce, apps that return completed work with citations attached across contracting, IP research, eDiscovery and regulatory risk. Sold as a fixed-fee engagement or annual subscription, it prices the output rather than the seat.
The Third Circuit ordered a lawyer to pay $14,636.50 in opposing counsel's fees after apparently AI-generated fictitious quotations appeared in a filing, and rejected the explanation that a law clerk's work had gone unchecked. The court's line was that misquotations are serious and counsel are expected to check citations, whatever drafted them.
AI took 60.3% of European VC deal value in H1 2026, up from 37.9% across 2025, or €26.5B of a €44B total. Deal counts fell while cheque sizes grew, so capital is available on a narrower thesis.
37% of law firm staff and 42% of their in-house counterparts report trouble fitting generative AI into existing systems, even as 92% use at least one AI tool daily, per Wolters Kluwer's Libra team. The remaining constraint is organisational: how work moves through a firm, and who checks quality at the end of it.
Pro se filings are climbing as self-represented litigants turn to AI, and judges in state and federal courts are answering with standing orders and sanctions when hallucinated citations reach the docket. Litigants without counsel are being held to counsel's verification standard.
Seventeen publishers, including The New York Times, are asking a federal judge to sanction OpenAI, alleging it deleted billions of ChatGPT output logs and obstructed discovery to hide how its models were trained. However the motion is resolved, how AI firms retain logs and preserve evidence is becoming a courtroom question rather than a back-office one.
Norm raised $120M at a $1.2B valuation for an AI law firm that runs its own agents under attorney supervision and bills clients on outcomes instead of hours. Capital is concentrating on legal AI that hands back finished work and prices it against the result.
Governance of AI-assisted legal work turns on proof that judgment was actually exercised, not merely on a lawyer's involvement. For legal and AI leaders, evidence of real human judgment becomes the standard to meet.
Copious launched Litigated.com, which extracts structured evidence from image-based PDFs and builds chronologies, disclosure reviews and claim assessments on infrastructure it controls rather than third-party cloud. Another entrant is staking out litigation-native workflow, with outputs still requiring review by a qualified lawyer before they touch a live matter.
AI director roles at major law firms pay up to $440k, yet the candidate pool barely exists as at least 16 firms, including Latham, Gibson Dunn, and Pillsbury, chase more than 25 hires. For legal-AI leaders, that scarcity sets a market price on a rare profile: technical enough to build and legal enough to be trusted.
Niche expertise and value-add beyond what general-purpose AI can generate is the differentiator for small and mid-size firms weighing build-versus-buy on legal AI, according to the National Law Review. For legal and AI leaders, that locates the advantage in domain-specific depth rather than in the underlying model.
The US government's temporary suspension of Anthropic's Claude Fable 5 is being used as a case study for why legal teams are now pulled into AI governance decisions at the enterprise level, not just contract review after the fact. For legal leaders, that signals an earlier and broader role in how AI gets governed across the business.
AI dealmaking reshaped Big Law's first-half M&A tables, with Gibson Dunn and Sullivan & Cromwell topping the league on SpaceX/xAI and OpenAI/Anthropic mandates. The work is concentrating at the top, which sharpens the pressure on smaller firms to compete on something other than scale.
HKU research finds law firms are adopting generative-AI workflows faster than they can govern them, with hallucination risk and unsanctioned 'Shadow AI' the open problems. Governance, not model capability, is what's lagging.
Courts are now sanctioning firms over AI-fabricated citations in filings, and the piece warns that opaque 'WL' (Westlaw) cites are especially easy to hallucinate and hard to verify. Left unchecked, a fabricated authority could work its way into the record, and even into precedent.
Kirkland & Ellis is co-building a custom platform with Palantir, and the operative word is ontology: a decision layer that sits over the data. The same idea scales down to a twelve-lawyer litigation boutique, where a finite universe of matters is small enough to model well.
In what's believed to be the first trial won by an AI lawyer anywhere, Garfield AI, the first SRA-regulated AI law firm, recovered a £7,000 debt for a client who paid it about £400 — beating an opponent that fielded both a solicitor and a barrister. At that price, contested debt and unpaid-fee claims that were too small to litigate economically become worth pursuing.
83% of in-house legal teams have no metrics to track whether their AI spend is paying off, per Axiom Law — only 17% can measure return at all. Budgets are climbing faster than the ability to prove value, which makes documented, measurable outcomes the next thing buyers will ask for.
Thomson Reuters' 2026 Future of Professionals report puts $143B of US client revenue at risk from the legal AI execution gap: 24% of talent now favour firms that deliver on AI, and a third of lawyers already use unsanctioned tools. The bottleneck is implementation, not capability, and the report's 'fiduciary-grade AI' bar (provenance, verifiability, a human in the loop) is the standard firms will be judged against.
Harvey is running proof-of-concept studies that train open-source LLMs on individual firms' workflows and client relationships, effectively encoding a firm's way of working into a model. The value is shifting from the base model a firm licenses to how well its own processes are captured — and a firm of any size can do that.
AI is rebuilding the economics of litigation on both sides of the 'v.', with plaintiffs and defendants alike now running it across discovery and drafting. The edge goes to whoever automates the grunt work first; the risk is being out-resourced by an opponent who already has.
In Artificial Lawyer's 2026 predictions, the LexisNexis chief executive forecast agentic systems running research, drafting and citation checking as one workflow, with traceable reasoning and human approval before anything is executed. Fully autonomous legal decision-making was what he said would not happen this year.