U.S. District Judge Sharion Aycock of the Northern District of Mississippi disqualified all four attorneys in a contract dispute this week after finding that both legal teams independently submitted AI-generated filings containing fabricated case citations — without coordination, and without verifying anything before filing. The hallucinations were scattered across three separate briefs submitted to the same judge. The case originated from a fee dispute between an attorney and the City of Aberdeen.

Judge Aycock's sanctions order did not spare the language. "In an era of rampant unverified AI usage within the legal field, this case presents a prime example of the risk associated with serving as a rubberstamp when acting as local counsel." Two out-of-state attorneys were banned from appearing in the Northern District for two years and fined $2,500 and $3,500 respectively. The two local counsel were disqualified from the case and ordered to pay $1,000 each. The matter was referred to the relevant state bar authorities.

Most of the conversation in legal AI is about the practice of law — drafting, research, diligence, contract review. Two funding announcements this week point to a different bet: that the business of running a law firm is just as large an opportunity, and far less crowded.

Billables AI — $10.2M Series A
Billables AI closed a $10.2M Series A led by Avenue Growth Partners, with Wing VC, SignalFire, and Alumni Ventures participating. The company passively captures attorney work activity, generates billing narratives automatically, and integrates with practice management platforms via API. It was named Legal Tech Company of the Year at the 2026 Legalweek Leaders in Tech Law Awards. The round announcement framed the ambition clearly: Billables AI is building toward "operational intelligence for law firms" — not just what was billed, but how work actually gets done, where time is lost, and what the firm's capacity and economics look like in real time. Law firms know almost nothing about their own operations in any structured way. Billables AI is betting AI can surface that data from the tools attorneys already use. If it works, it becomes the kind of infrastructure a firm can't turn off.

Sandstone — $30M Series A
On the same day, Sandstone announced a $30M Series A led by Lightspeed Venture Partners, six months after a Sequoia-led seed. Sandstone is building AI infrastructure for in-house legal departments at small and mid-sized companies — not law firms. The positioning is deliberate. Harvey, Legora, and the major legal AI platforms have concentrated almost entirely on private practice. In-house teams have different problems: they live in Slack and email, not document management systems; they need approval workflows, not drafting tools; and they are often a team of one or two managing legal work across an entire company. Lightspeed's bet is that this market is large enough to support a dedicated platform — and that the firms fighting over BigLaw are leaving it entirely uncontested.

Anthropic released Claude Fable 5 this week, and the legal results are the most notable thing about it. On Harvey's open-source Legal Agent Benchmark — 1,200 realistic tasks across 24 practice areas, graded against expert rubrics where missing a single criterion means failing the task — Fable 5 now sits at the top of the leaderboard with a score of 13.3%, up from Opus 4.8 which was the first model to crack 10% just weeks ago.

Mark Pike, who leads Claude for Legal at Anthropic, shared early testing results publicly. Crosby, a legal AI company, ran Fable 5 head-to-head against their current model in blind review. The finding: their lawyers found Fable 5's redlines matched or beat their current model every time. Pike described it as feeling "materially different."

The number that matters most is not 13.3% in isolation — it is the rate of improvement. A few weeks ago the ceiling was under 10%. It is now 13.3%. If foundation model performance on complex legal tasks continues improving at this pace, the question of what firms are actually paying Harvey and Legora for — versus what they could get from a foundation model directly — will become harder to answer. Harvey's moat has always been workflow, integrations, and firm relationships, not the underlying model. But every time the model gap closes, that argument has to work harder.

Spellbook announced a fellowship program this week offering law students $25,000 in funding, mentorship, and full access to the platform and beta features to build legal tech ideas. The pitch is direct: come up with something that meaningfully reduces friction in day-to-day legal work, and Spellbook will back it.

The announcement is small in dollar terms but points at something larger. Law schools have historically been the last place anyone looked for legal tech innovation — the culture runs toward doctrine, not product. That is changing. As AI makes it easier for non-engineers to build functional tools, attorneys who understand the actual problems are increasingly the ones best positioned to solve them. Spellbook is betting that the next wave of legal tech ideas lives inside law school classrooms, not Silicon Valley pitch decks. They may be right.