Legora crosses $100M ARR. A Georgia prosecutor is suspended over AI-fabricated citations. The White House publishes its AI preemption framework. Law360 finds 70% of attorneys now use AI weekly. And the legal AI arms race enters a new phase.
The combined ARR of just two companies — Harvey and Legora — now exceeds $300 million. The legal AI market has moved from interesting experiment to infrastructure spend in under two years.
Legora's growth trajectory has been extraordinary by any standard. Bessemer Venture Partners reported that the company reached $100M ARR faster than OpenAI, Anthropic, Cursor, and Wiz. Investors who priced the company at $5.55 billion in its Series D were effectively paying 240x reported ARR at the time. With the new revenue disclosure, that multiple drops to roughly 55x — still aggressive, but within the range the market has accepted for high-growth vertical AI businesses.
Harvey, meanwhile, confirmed it passed $200M ARR in late March. The gap between the two companies is narrowing, but Harvey remains deeply embedded among the Am Law 100 and has a meaningful head start in enterprise deployments.
Both Harvey and Legora sell to law firms — helping expensive resources do work slightly faster. Neither is competing for the outsourcing budget. Neither is replacing the human in the loop. The firms paying for these tools are capturing efficiency as margin, not passing it through as price reductions to clients.
The more interesting question isn't which legal AI copilot wins. It's when the buyer shifts from the firm to the in-house team — and whether the winning product at that point looks like a research assistant or a supervised agent that does the work end-to-end.
The Georgia Supreme Court incident is the highest-profile AI citation scandal since the Mata v. Avianca case in 2023 — but it's no longer an anomaly. According to Artificial Lawyer, documented cases of AI-hallucinated citations in legal filings are now being discovered at a rate of four to five per day, up from 120 total cases between April 2023 and May 2025 to over 660 by December 2025.
The Georgia case is notable for three reasons: it involved a prosecutor rather than a civil litigant, it reached the state's highest court, and the fabricated citations were embedded in a proposed order denying a murder conviction appeal — meaning they could have directly influenced the court's ruling.
Every one of these incidents follows the same structure: a human treats AI output as a draft that merely needs formatting, rather than as raw material that needs verification. The tool isn't the problem. The absence of a supervision layer is. Courts are now moving toward mandatory hyperlink rules that would require every citation to link to a verified legal database — a mechanical fix for a workflow problem.
The White House's March 20 framework represents the administration's most detailed attempt yet to establish a national AI regulatory standard. The core proposition: replace the growing patchwork of state AI laws with a single, "minimally burdensome" federal regime.
The framework addresses seven policy areas — child safety, community impacts, copyright, free speech, federal regulation, workforce, and state preemption — but its most consequential recommendation is the push to prevent states from regulating AI model development or imposing liability on developers for third-party use of their systems.
| Regulation | Jurisdiction | Effective date | Relevance |
|---|---|---|---|
| Colorado AI Act | Colorado | June 2026 | Risk management, impact assessments, and transparency for high-risk AI systems. Colorado's working group is proposing a lighter replacement framework before it takes effect. |
| EU AI Act (high-risk) | European Union | August 2026 | Full application to high-risk AI systems. Legal services fall within scope. Penalties up to €35M or 7% global revenue. Conformity assessments required. |
| Oregon SB 1546 | Oregon | January 2027 | Chatbot operators must implement protective measures for minors. Signed into law March 31, 2026. |
| Take It Down Act | United States | May 2026 | Prohibits nonconsensual publication of intimate visual depictions, including AI-generated deepfakes. Platforms must remove upon notice. |
The most interesting finding in the Law360 Pulse survey isn't the adoption rate — 70% weekly use was predictable given the trajectory. It's the sentiment shift. Among attorneys who use AI three or more times a week, positive sentiment has dropped from 73% to a minority position, with 44% now describing their view as neutral.
This looks like maturity, not disillusionment. Early adopters who expected transformation are discovering that the tools are genuinely useful for acceleration — research, summarisation, first drafts — but don't fundamentally change the work. The cognitive load doesn't disappear. It shifts. And the hourly billing model means that time savings often compress revenue rather than create capacity for higher-value work.
The adoption numbers are a lagging indicator. What matters now isn't whether lawyers are using AI — they are. It's whether the delivery model is changing. If 70% of attorneys are using AI weekly and the billable hour is still the primary revenue unit, the technology is being absorbed into the existing structure rather than disrupting it. The disruption comes when the buyer — the in-house team — starts purchasing outcomes instead of hours. That's the shift from AI-as-tool to AI-as-production.
The legal AI market is bifurcating. On one side: tools that help lawyers work faster. On the other: systems that do the work, with lawyers supervising. This week's news illuminates both paths — and the gap between them.
The copilot market is maturing fast. It's well-funded, well-adopted, and increasingly commoditised. The more interesting structural question is what happens when enterprise legal teams stop buying tools that make their outside counsel faster — and start buying systems that replace the outside counsel entirely for routine work.
That's the difference between a $500 research tool and a supervised agent that handles your NDAs, MSA redlines, and procurement contracts end-to-end — for a fraction of the cost. The budget for that isn't software. It's the external legal spend line.