News You Can Use

For two years Anthropic was the upstream - the model that Harvey, Legora and CoCounsel built on. On 12 May they shipped "Claude for Legal". This is more than 20 MCP connectors into legal systems plus 12 practice-area "plugins", each set up through an interview that captures a team's playbooks, escalation chains, risk calibration and house style. Four of them - Commercial, Corporate, Litigation, Product - ship as managed agents through the Claude API. The whole thing runs across Microsoft 365 as a single context-carrying assistant, so the matter you are working on follows you from Word to Outlook to Excel. Freshfields, Quinn Emanuel, Holland & Knight and Crosby Legal were named as live users. Harvey, Thomson Reuters, Relativity and Everlaw all turned up as integration partners rather than competitors. There is clearly demand (more than 20,000 people registered for Anthropic's "Claude for Legal Teams" webinar in April.) and is an intentional move into legal as a vertical, in a similar way they moved into Finance recently.

The bet is that the primary interface for a lot of legal work becomes Claude inside Word, and everything else - the DMS, the data room, the e-discovery platform, the research stack - becomes a connector feeding it rather than a destination you log into. The application layer stops being the place you go and becomes a set of sources. The impact is uneven though. It is not especially hard for Harvey or Legora; firms don't use those tools for model access but for purpose-built adoption and change management. The layer this is actually hard for is the thin one below them, the "LLM with a legal skin" tier. The market priced the launch the same way: Thomson Reuters closed -6.1% on launch day on a beat-and-raise quarter; RELX and Wolters fell too. The market has decided disintermediation is the dominant story, but TR's actual AI business is growing fast. The question Claude for Legal forces, is no longer "which legal AI vendor" - it is "what do we still need a vendor for?" For BigLaw it's all the behind the scenes work that a serious innovation function is for but for a lot of mid-market and in-house teams that already have Claude Enterprise, the answer is "not much". The floor has moved up, the model is the commodity, the drafting surface is contested ground, and the work that survives is the work that needed a human with judgement in the first place.

Same fortnight as Claude for Legal, the model providers walked into the implementation business too. Anthropic announced a $1.5bn enterprise AI services joint venture with Blackstone, Hellman & Friedman and Goldman Sachs - forward-deployed engineers Palantir-style, embedded inside PE-portfolio companies and mid-market clients, redesigning workflows around Claude. The same day, OpenAI announced its a $10bn JV anchored by TPG with Brookfield, Advent and Bain Capital, branded "The Deployment Company". Anthropic shipped the vertical version of the same play for finance - ten ready-to-run Financial Services Agents, Claude Opus 4.7 tuned for finance, and Moody's data integration. Microsoft completed the platform picture with Agent 365 hitting general access and Copilot Cowork shipping (US only). The application layer ran the same play in miniature, with Harvey shipping 500+ pre-built agents plus a "Transformation Office" of ex-BigLaw, and a PwC case study. And on the law-firm side, Linklaters launched Applied Intelligence - a productised team pairing lawyers with data scientists, multi-frontier-model by design.

The model providers are entering the implementation business directly with PE money behind them, and the competitive surface area is moving. This threat is implementation-layer disintermediation a downstream move by model providers. For buyers, the procurement question focuses on the team and the implementation approach, rather than the models. Forward-deployed engineers paid by an Anthropic-Goldman vehicle do not have the same alignment as independent advisors paid by you. The approach from Linklaters is to productise the lawyer-plus-data-scientist team and sell it to clients. We are seeing the model providers, Big Tech, Legal Tech and law firms all moving into the same adjacent territory. For independent consulting, the gravitational pull on services budgets is now visible and the honest concern is not "AI replaces consultants" - it is that an Anthropic-Goldman JV with $1.5bn of underwritten capacity can compress the price of the implementation layer faster than independents can repackage their offer. The answer may be to be the team that knows the firm and the work well enough to direct the FDEs rather than be replaced by it.

While the model providers were coming downstream, a new cohort of AI-native firms walked upstream into legal services itself, with the beginnings of a new category. Carta (the $7bn private-capital ERP) acquired UK ABS Avantia and rebranded as Carta Law. Talairis Law Group launched in Seattle. LawFairy is back in the news as the second SRA-authorised tech-only UK firm and the first deterministic, non-generative-AI firm to receive approval - rule-based "FairyLogic" across 55+ UK immigration routes. Moritz closed a $9M Series A in four days, reporting 100+ companies served, ~$2bn aggregate contract value, four-hour average turnaround across commercial / corporate / employment. Behind them sit Manifest OS ($60M Series A at $750M, US business immigration), Keith (UK conveyancing, CLC-authorised, £2M seed), the existing Crosby / Lavern / Farringdon / NewMod / Garfield cluster, and Brazilian entrant Enter (valued $1.2bn). Eudia bundled "Expert Digital Twins" of senior lawyers into a unified in-house workspace.

Six funded AI-native entrants in two weeks, on top of an existing cluster, in a market where new commercial law firm formation has been a non-event for two decades. Each one focuses on a different category - client type, jurisdiction, practice area, even technique. Each area is narrow enough to defend against an incumbent with broader scope and slower price discipline, and most are pricing on outcome, fixed fee or pay-as-you-go rather than the hour. The right read is not "AI replaces BigLaw" - it does not, not at the work BigLaw is paid for, not soon. But for any defined slice of legal work, a verticalised AI-native firm can now stand it up with a credible regulatory perimeter, a defensible quality story, and pricing the incumbent can't match without cannibalising its own model. For the UK specifically, two SRA tech-only authorisations (Garfield; LawFairy), one CLC-authorised conveyancing entrant (Keith), and Magic Circle interest in the Carta deal say the regulator has settled on "yes, with conditions" as its working position - and the deal flow is responding. The strategic question for a BigLaw firm is which slices of its income are inside the cohort's blast radius and which are not. And that perimeter isn't fixed. BigLaw is broadly safe today because most of its book still needs human judgement and bespoke handling that a vertical entrant can't replicate. But the same tech the firms are rolling out internally is what commoditises the next slice of work, and a commoditised slice is one a vertical entrant can stand up. The defence holds for now; it gets harder every cycle.