OpenAI launched Astra for Law on September 17, selling a legal-tuned GPT-6 Astra directly to large firms while continuing to license the same model to Harvey and Legora. New BigHand research found 63% of law firms still price matters by habit, standard hourly rates or a copied past invoice, even as 56% of their clients demand AI-driven efficiency. White & Case took a stake in a Saudi contract-review platform built by its own former Riyadh partner, and Clio hired former Michigan Chief Justice Bridget Mary McCormack to help sell AI tools into the courts.
OpenAI released Astra for Law on September 17: a configuration of GPT-6 Astra paired with a legal search index covering US case law, statutes, regulations and administrative decisions, refreshed daily, built with the nonprofit Free Law Project to cover more than 99.9% of published US precedential case law. The product reaches selected Am Law 200 firms directly through Trusted Access in ChatGPT, an access tier OpenAI is designing with Latham & Watkins around information permissions and ethical walls, while Harvey, Legora and other vendors keep the same API relationship they've always had.
On the private validation set of Vals AI's Legal Research Bench, Astra for Law answered 54% of 200 legal research questions correctly, against 38.7% for the base GPT-6 Astra using web search alone, OpenAI reported at launch. Pricing for the legal configuration hasn't been published. For firms it reaches, the offering includes zero data retention on the API and excludes ChatGPT Enterprise usage from human review by default, per OpenAI. Harvey's head of applied research, Niko Grupen, called the underlying model "a significant quality improvement" on tasks like spotting unsupported assumptions in a document, Artificial Lawyer reported September 7. One technologist quoted by Legal IT Insider tied that same capability jump to the disappearance of the file-review work that has traditionally trained junior lawyers: "We're now moving to the level of agentic AI that removes junior training. We knew it was going to happen."
OpenAI now supplies the model that Harvey and Legora price their software on top of, and sells a version of that same model directly to the firms that might otherwise buy Harvey or Legora instead. Whoever wins that fight, a faster research tool doesn't replace the part underneath it: deciding which matter goes to which resource, and making sure a person checks the output before it reaches a client or a court. A 54% score on a research benchmark is a starting point, not a finished answer. Something still has to verify it, and that's inexpensive work landing on an expensive resource until a legal team builds the layer that routes and reviews by default, regardless of which lab's model is running underneath.
BigHand's 2026 Legal Pricing and Budgeting Trends Analysis, published September 15, found that 63% of law firms still set a matter's price the way they always have: 35% default to a standard hourly rate, and another 28% copy the number from a past, similar matter without adjusting it. Only 1% of firms report using a template built into a pricing or budgeting tool, the mechanism that would actually let a lower AI-driven cost show up in the number a client sees.
More than half of firms, 56%, say clients are already asking for AI-driven efficiency or greater transparency about how AI use affects a bill, according to the report covered by Legal IT Insider and Legal Futures. Firms have invested in pricing expertise, BigHand notes, but that insight often reaches a partner only after a price is already set, not before it. Separately, 35% of firms cite partner discomfort discussing AI with clients as the single biggest barrier to having that conversation at all.
This is inexpensive work done by expensive resources, showing up in finance data instead of case law. A partner still prices a redline or an NDA off a rate card built for the years before AI, because nothing in the firm's own pricing tool knows the work now takes a fraction of the time it used to. Routing that work to supervised agents doesn't just change who does it. It forces the pricing conversation BigHand's own numbers say isn't happening: what a matter actually costs to deliver today, not what it cost three years ago.
White & Case confirmed on September 10 a strategic investment in Clauze.AI, an AI contract review and due diligence platform built for legal and corporate clients in Saudi Arabia and the wider Gulf region, Artificial Lawyer reported. The firm didn't build the tool itself. It backed one built by Waad Alkurini, who spent a decade at White & Case, most recently as executive partner of its Riyadh office, before leaving to found the company.
Clauze.AI is built around the region's specific requirements: bilingual Arabic and English review, full data residency inside Saudi Arabia, and on-premises deployment for clients whose data can't leave the country, according to the company's own launch materials. It's White & Case's first investment in a Middle East technology company. The terms, including the amount invested, haven't been disclosed.
A law firm investing in the AI tool its own former partner built, rather than building the equivalent capability under its own roof, is a small vote for the same thesis wherever it shows up. Reviewing a contract against a team's own playbook and flagging what needs a human's attention is routing infrastructure, and routing infrastructure is worth buying, not worth every legal team reinventing separately. That's true whether the client sits in Riyadh or wherever a legal team is routing its own NDAs and redlines. The build-versus-buy question hasn't changed. It has just travelled to a new market, with a new regulatory reason, data residency, for the answer to keep landing on buy.
Clio announced on September 16 that Bridget Mary McCormack, the former chief justice of the Michigan Supreme Court, will join the company in October as general manager, judiciary, alongside Casetext co-founder Pablo Arredondo. McCormack is leaving the American Arbitration Association, where she has served as president and CEO since 2022, effective October 2, to help Clio sell into the courts, judicial officers and dispute resolution bodies she spent her career inside.
"Courts have been operating under structural pressure for years, and technology now gives us a chance to rethink what support can look like," McCormack said in Clio's announcement. Clio CEO Jack Newton called the judiciary "one of the largest opportunities to strengthen how the legal system works." McCormack will lead market and commercial strategy for Clio's judiciary business; Arredondo leads its product and strategic direction.
McCormack ran the country's largest arbitration provider and, before that, sat on a state's highest court. Whether that background produces genuinely better court technology or simply lets one vendor set the defaults judges and clerks get used to isn't something a press release settles. If Clio's judiciary tools end up shaping how a court schedules your matter or reviews a filing, whose interests were in the room when those defaults were designed is worth asking before your team assumes the answer is yours.
OpenAI decided it doesn't need Harvey or Legora to reach a law firm. BigHand's data shows most firms haven't repriced their own work to reflect what AI now makes possible. A law firm bet on a founder who used to work for it, rather than building the tool itself. And a former chief justice is about to help one vendor shape how courts use AI. None of these four stories is really about better AI. All four are about who gets to decide where the work goes, and who profits once the work is done.
Take these four together: a model maker skipping the software layer it used to sell through, a pricing model that hasn't moved, a law firm buying instead of building, and a vendor recruiting the person who used to run the country's largest arbitration body to now sell into courts. None of it changes the actual bottleneck sitting inside a legal team today.
Outsource legal work to supervised agents, and the question stops being which lab's model a vendor licenses this quarter. It becomes a mechanism: agents that draft and route against a team's own templates, terms and escalation rules, with a lawyer reviewing the output before anything leaves the building. Inexpensive work stops being done by expensive resources, whichever company is selling the model underneath.