Welcome to Issue #1 of The Play — a weekly brief for legal ops directors, law firm admins, and GCs figuring out AI before their competitors do. Every issue is built around plays — moves, reads, and decisions worth making. Not trends to file away for 2027.
Here's what's on my radar.
The lawyer who gets sanctioned next isn't using ChatGPT. They're using a legal-grade AI tool — Westlaw integration, enterprise contract, the whole thing. The Ninth Circuit just made clear that none of it matters.
THE LEAD PLAY
Your AI Tool Doesn't Own the Liability. Your Lawyer's Signature Does.
Three weeks ago, the Ninth Circuit issued a published, precedential opinion in LNU v. Blanche, sanctioning two attorneys for AI-fabricated citations. Published. Precedential. This isn't a district court experiment. It's binding circuit law, and every other circuit is taking notes.
Then, effective June 15, Florida's Supreme Court amended Rule 2.515 to require the signer of any filing to certify that all legal authorities cited "exist and are accurately cited." The rule applies to every filing in Florida state court. Available sanctions: reprimand, contempt, dismissal, attorney's fees.
Taken separately, both are significant. Taken together, they establish something the legal ops world hasn't fully absorbed yet.
The liability doesn't live in your AI policy. It doesn't live in which tool you approved, or what the vendor's marketing says about accuracy, or whether your enterprise plan includes a citation-checking feature. It lives at the moment a lawyer puts their signature on a document. That signature is now a certification. If a fabricated case makes it through your review process and into a filing, no amount of enterprise security features or "trained by lawyers" copy changes what happens next.
We're at 363+ sanctioned matters on the docket. A Mississippi federal court removed all four attorneys in Withers v. City of Aberdeen last week — both sides, fabricated citations. That one is going to get attention.
The Play this week: Adopt a written, tool-agnostic citation-verification protocol and circulate it before end of week. Not a ChatGPT-only policy. A protocol that covers everything in your drafting stack — Harvey, CoCounsel, Perplexity, whatever's in rotation. It needs two steps: (1) a deterministic check that a citation actually exists at the cite given, and (2) a read-the-source confirmation that it says what the brief claims it says. These don't need to be automated. A junior associate with 10 minutes and Westlaw access can run both.
If you have litigators in Florida or matters in the Ninth Circuit, brief them on the new certification language now. Then check your courts' standing orders — more than 100 federal courts have AI disclosure requirements on the books, and your attorneys probably don't know which of their courtrooms are covered.
The firms that get sanctioned next won't be the ones with no AI policy. They'll be the ones whose policy stopped at "approved tool list."
SUPPORTING PLAY 1
The $143B Wake-Up Call. Also: You Probably Don't Have a Scorecard.
Thomson Reuters released its fourth annual Future of Professionals report this week. The headline number getting attention: a modeled ~$143 billion in U.S. legal and accounting revenue that corporate clients will be actively reconsidering within 12 months, based on which providers can demonstrate AI-driven value.
That number is vendor-modeled, not independently audited. But the underlying survey data — 1,816 professionals, 62 countries — is harder to dismiss. 74% of legal professionals say they use AI tools several times a week. Only 18% of organizations actually measure whether it's working.
That gap is the real story. Clients are already asking. In-house legal teams are comparing their AI productivity against outside counsel — and some, like Ford's GC this week, are now saying publicly that their internal adoption has outpaced their law firms'. When your top client asks "what are you doing with AI" in the next quarterly review, "we're exploring it" isn't a credible answer anymore.
The Play this week: Stand up a minimal ROI scorecard. You don't need a dashboard. You need three numbers you can speak to with confidence: (1) time saved per workflow — pick your highest-volume one and start there, (2) AI adoption rate among attorneys, and (3) client-mandate compliance — how many clients have asked about your AI use, and do you have a documented answer. Track them monthly. The data matters less than having a credible, consistent story when the question comes up.
While you're at it: survey your staff on unsanctioned AI use. The report puts shadow AI at roughly one-third of legal professionals. That's not a guess — it's a benchmark. Worth knowing where you actually stand before a client asks about your governance.
SUPPORTING PLAY 2
The Faegre Drinker Template. Steal It.
Faegre Drinker deployed Harvey and Microsoft Copilot across all ~1,200 attorneys and consulting professionals this week. Not a pilot. Not "access for interested attorneys." Firmwide.
Two things make this worth paying attention to. First, the structure: a vertical legal tool (Harvey) for the work product layer — drafting, diligence, research — plus a horizontal tool (Copilot) for the Microsoft 365 layer where the actual workflow lives — email, Word, Teams. Two layers, two tools, each doing what it's actually good at. Most firms either pick one or let the tools sprawl without a framework.
Second, the training: mandatory, role-based AI ethics training before anyone got broad access. Not optional. Not a lunch-and-learn. A prerequisite. That piece is what makes the rollout defensible — to clients and to courts.
The Play this week: Map your current stack to this two-layer model. Do you have a legal-specific vertical tool for work product? A horizontal tool for the productivity layer? And a documented, role-based training requirement before access is granted? If you're missing any of those three, that's your gap. The training piece especially — Faegre Drinker treated it as a gate, not an afterthought. That distinction matters more than it used to.
QUICK HITS
Perplexity launched "Computer for Counsel" on June 24. An agentic system routing across 20+ AI models and connecting to tools legal teams already use — Box, DocuSign, Ironclad, NetDocuments, and others. For mid-size firms priced out of Harvey, this is the one to watch. Gunderson Dettmer, which went firmwide with Perplexity last year, reports 80% attorney adoption and 35,000+ queries per month. Some integrations are still listed as "coming soon" — pilot before committing budget.
Clio is training 25,000 legal professionals for free. The Legal AI Accelerator runs through August 31 — 65 days, self-paced, CLE-eligible. If baseline AI fluency is the bottleneck on your team (it is), enroll associates and staff now. Zero cost. The Florida Bar separately made four months of Clio's AI workspace free for eligible members — the first state bar to do this. Watch for others to follow.
Kirkland confirmed a ~$500M build-your-own-AI commitment (~$100M in 2026 alone). The Am Law 1 signal: at the very top of the market, the answer is "build, not buy." For everyone else, vendor tools continue to commoditize. Kirkland's custom stack will be a meaningful competitive differentiator within three years — and a useful data point when your managing partner asks why you're not just building something too.
Harvey power users now save an average of 11 hours per week, up from 8.5 six months ago, per RSGI's second annual Harvey report (87 respondents, 60 firms, 27 in-house teams). 89% say they can take on more work without adding headcount. Harvey-adjacent research — treat the numbers as directional. But the trajectory is consistent with what operators are reporting internally, and the delta from six months ago is the part worth tracking.
That's Issue #1. Next week: what a functional AI governance policy actually looks like — not the 40-page document nobody reads, but the three-page version your attorneys will actually follow.
See you then.