The Lab Notebook: No. 5
Welcome to The Lab Notebook, VAILL’s field log for tracking how AI is reshaping legal work.Our goal: fewer hot takes, more meaningful signals. We’re designing each Notebook entry to reflect an industry in motion: Version Control highlights recent meaningful updates, Open Query surfaces the unknowns we should be debating, Network Effects highlights what’s compounding through shared practice, and System Status checks the guardrails and institutional infrastructure. We close with The Signal, one high-fidelity takeaway that suggests where the future is heading.
🔁 Version Control
The model providers walk into Word. Anthropic released Claude for Word in public beta on April 10; Microsoft followed three weeks later with its own Legal Agent in Word, built by the team that came over from Robin AI. Both led with legal contract review as the flagship use case. Anthropic · Microsoft
Lawyers go DIY in public. California attorney Helen Fan is running OpenClaw Law LLP publicly on LinkedIn with two AI agents; her May 1 post described her senior agent autonomously generating its own practice manual after just three prompts. And just days ago, ex-Latham & Watkins associate Will Chen released Mike OSS, an open-source self-hostable legal AI platform he built in two weeks using Claude. The largest commercial platform in this category is valued at $11B. Former Leigh Day IT Director James Harrison’s verdict: “Once a working open-source alternative is sitting on GitHub, the conversation in renewal meetings moves from ‘Is this magic?’ to ‘What exactly am I paying enterprise prices for?’” Helen Fan (LinkedIn) · Artificial Lawyer (Mike) · Legal IT Insider
VAILL take: Is the legal tech vendor layer being bypassed from both ends? Model providers reach past it into the lawyer’s working surface; lawyers and ex-lawyers reach below it to build their own. Capability is increasingly available without the platforms that have historically sold it.
🧐 Open Query
Whose policy governs the agent? As agents move from chatbot to substantive workforce, Rule 5.3 supervisory duty and ABA Formal Opinion 512’s verification requirement both run into bandwidth limits. When an agent’s autonomy outpaces the lawyer’s effective oversight, where does the obligation truly rest?
Verify the verifier? On April 27, E.D. Pa. Judge Kai N. Scott sanctioned attorney Raja Rajan $5,000 — after a prior $2,500 sanction in the same case — for filing a brief written by one AI chatbot whose citations he asked a different AI chatbot to verify. Six fabricated citations made it through. Under what conditions, if any, can AI-generated verification satisfy a lawyer’s duty to confirm cited authority? Bloomberg Law
VAILL take: Both questions sit in the gap between deployed capability and reviewable oversight. The next round of court orders and ethics opinions will have to define oversight at speeds capability now exceeds.
🌐 Network Effects
The firm becomes the defendant. In an early April order in Rivera v. Triad Properties Corp., N.D. Ala. Judge Anna Manasco sanctioned both attorney Joshua Watkins and his former firm Burrill Watkins LLC under Rule 11 — the firm because it could not show how it enforced any AI policy. Manasco extended her Johnson v. Dunn doctrine from individual attorneys to the firm itself. Also, $47K in sanctions. Ouch. Reason / Volokh Conspiracy
DOJ joins the docket. On April 28, E.D.N.C. Magistrate Judge Robert T. Numbers II formally reprimanded former Assistant U.S. Attorney Rudy Renfer for filing an AI-generated brief with fabricated quotations and for initially failing to disclose his AI use under questioning. Numbers limited the sanction to public reprimand only noting Renfer had already been fired before he could resign. ABA Journal
The price keeps climbing. Federal courts in Oregon imposed approximately $109,700 against two attorneys for two dozen fabricated citations under a per-citation formula. The Nebraska Supreme Court issued what appears to be the first indefinite license suspension tied to AI hallucinations in a court filing. NPR
VAILL take: Issue No. 4 noted that 9% of firms surveyed had a written, actively enforced AI policy. Manasco moves that statistic from a governance-lag indicator to a litigation risk; Renfer adds that institutional employers like DOJ provide no shield once conduct lands before a court. The cost of “no policy” has changed — and so has the cost of a policy that exists on paper but does not work.
🚦 System Status
Federal preemption acquires a draft. On March 20, the White House released its National Policy Framework for AI — a four-page legislative blueprint endorsing federal preemption of state AI laws that impose “unduly burdensome” requirements. Two days earlier, Sen. Marsha Blackburn (R-Tenn.) circulated a 291-page draft of her TRUMP AMERICA AI Act, codifying elements of December 2025’s “One Rule” Executive Order. White House · Holland & Knight
California writes AI into the rules. On March 13, the California Bar’s Standing Committee on Professional Responsibility and Conduct approved amendments to six Rules of Professional Conduct that would embed AI obligations directly into enforceable disciplinary rules. Among them: a Rule 1.1 comment requiring lawyers to “independently review, verify, and exercise professional judgment regarding any output” of AI used in client representation, with no carve-out for routine matters; and a Rule 5.1 amendment requiring managerial lawyers to establish AI-governance procedures. The 45-day comment period closed May 4. LawSites
VAILL take: States and bars have done most of the legal-AI governance work to date — ethics opinions, 300+ federal standing orders, last issue’s Oklahoma rule split. California would now move from persuasive guidance to enforceable rules; what’s the interplay (if any) with the federal Framework and Blackburn draft? Helen Fan’s California experiment would run directly into Rule 1.1’s verification requirement and Rule 5.1’s firm-policy mandate.
📡 The Signal
Rule 11 finds the lawyer first. Always has. Now, it finds the firm too. And the vendor that used to absorb some of the diligence is fading in the age of open source.
The firm is on the hook. Judge Manasco sanctioned Burrill Watkins LLC alongside the named attorney for failing to enforce its own AI policy.
Capability arrived without a vendor. Helen Fan ran an AI-native firm in public. Will Chen launched a potential challenger to an $11B platform in two weeks. Anthropic and Microsoft put model agents inside Word.
The price climbed. Oregon imposed roughly $109,700 in sanctions under a per-citation formula. Nebraska suspended a license. Rudy Renfer’s seventeen years at DOJ did not stop a court reprimand — the office had already fired him.
What lawyers used to delegate to vendor vetting, they and their firms now answer for directly.