ADMISSIBLE AI ISSUE 001
The week in legal AI, explained in plain English · September 15, 2026 · 4-min read
A murder appeal quoted witnesses who never existed. The court fined the lawyer $5,000 — and sent his name to the disciplinary board.
September 2026 Edition · The legal-AI newsletter for people who bill by the hour
The penalties are getting specific. This week New Mexico's top court fined a lawyer $5,000 over witnesses who never testified. A Los Angeles court fined an associate $999.99 over citations that do not exist. Meanwhile 94% of law firms say they use or explore AI. That gap is the issue.
The big picture
AI adoption is settled: 94% of law firms now use or explore generative tools. What this week settled is who answers when it goes wrong. Courts priced two failures — a $5,000 contempt fine in a murder appeal, $999.99 for fake citations in Los Angeles. A securities suit survived dismissal because AI's role in the numbers was never disclosed. A regulator told firms to review how staff use the tools. And a vendor wrote its models a rulebook that outranks your prompts. The question moved from 'do you use AI?' to 'can you prove what it did?'
⚡ THE 30-SECOND BRIEF
1 New Mexico fined a lawyer $5,000 for AI-invented testimony in a murder appeal.
2 Associate fined $999.99 for fake citations, then told the court how she will verify.
3 Morgan & Morgan will spend $1 billion on AI — and sell its platform.
4 A judge let a securities case over AI disclosure against CVS proceed.
5 94% of law firms use or explore AI; Microsoft Copilot leads by a mile.
6 FINRA firms are told to review how employees actually use AI.
7 Microsoft wrote its models a rulebook that overrides what users ask.
WHAT'S ACTUALLY CHANGING
The seven stories that matter this week
1 Court Fines Lawyer $5,000 for the Witnesses AI Invented

Above the Law (via Reuters) · Sep 14 · relevance 28/30

New Mexico attorney Stephen Aarons filed an appellate brief in a murder case using testimony from witnesses who never existed. He asked ChatGPT to summarize the trial transcript. He assumed it returned 'a bulletproof summary.' The state Supreme Court held him in contempt. It fined him $5,000 and found 'a lack of remorse and a lack of concern for his client.' Justice C. Shannon Bacon framed the week with a blunt question: 'Counsel, do you watch the news?' Translation: AI no longer only invents case law — it invents evidence, which is much harder to spot.

WHY IT MATTERS

If you file briefs, read the record yourself before citing it. A fabricated witness is a disciplinary referral, not a typo.

2 An Associate's Apology Came With a Verification Plan

ABA Journal · Sep 15 · relevance 27/30

Jacquelene Robinson, a senior associate at Musick, Peeler & Garrett, filed a brief with fabricated case citations while defending State Farm in a contract dispute. Opposing counsel caught it, and the Los Angeles County Superior Court fined her $999.99. In a September 3 declaration she apologized and named her fixes. She will pull every authority from Westlaw or LexisNexis, check each quote against the real opinion, and audit citations before filing. She also took a CLE course on AI ethics. Translation: the useful part of a sanction is the procedure it produces.

WHY IT MATTERS

Copy that checklist into your own filing routine. It costs an hour and removes the easiest way to get sanctioned.

3 Morgan & Morgan Bets $1 Billion, Then Sells the Platform

ABA Journal · Sep 15 · relevance 27/30

Morgan & Morgan will invest $1 billion in AI and technology over the next decade. It also plans to sell MX2, its platform for extracting medical records and drafting case documents, to other firms by the end of 2027. The firm spent $300 million building MX2, which now has nearly 5,000 monthly users. A 2025 sanction hangs over the firm for citing eight nonexistent cases. Chief transformation officer Yath Ithayakumar said it happened 'because we were early adopters of AI,' and that training has improved since. Translation: the largest plaintiff firm in America is now also a software vendor.

WHY IT MATTERS

If you buy legal AI from another firm, ask who verifies the output. Their mistakes can land as your malpractice exposure.

4 CVS Must Answer for What Its AI Never Disclosed

LexBlog (D&O analysis) · Sep 15 · relevance 26/30

On August 27, a federal judge in the Southern District of New York let part of a securities class action against CVS continue. The surviving theory is not about AI capabilities. It says CVS described the drivers of its financial performance while omitting how AI-assisted prior authorization processes contributed to cost savings and profitability. The court rejected claims built on CVS's 'responsible AI' language as corporate puffery. CVS filed a motion for reconsideration on September 10. Translation: the risk is no longer what you say about AI — it is what you leave out.

WHY IT MATTERS

If AI touches your reported numbers, ask whether investors are entitled to know its role. Silence is now the claim.

5 What Law Firms Actually Use: Copilot, by a Wide Margin

LawSites · Sep 14 · relevance 26/30

ILTA surveyed more than 500 law firms, from small boutiques to global firms. Ninety-four percent say they use or explore generative AI, up from 80% a year ago. The most-used tool is not a legal AI product: Microsoft 365 Copilot, at 76%. CoCounsel (44%), Claude (44%) and Harvey (43%) are effectively tied behind it, with Legora at 16%. Only Copilot and Westlaw Advantage are fully deployed across at least half of firms' lawyers. Translation: most AI in law today is a general-purpose assistant bolted onto mail and documents.

WHY IT MATTERS

Benchmark your stack against the survey before your next renewal. If you buy legal-specific AI, know that most peers have not.

6 FINRA Firms Told to Review Employees' AI Use

LexBlog (Greenberg Traurig alert) · Sep 15 · relevance 24/30

A new client alert says FINRA member firms are deploying approved AI tools quickly, and that employee use will grow just as fast. Most people will use them properly. Firms should still expect misuse, workarounds of existing controls, and compliance failures nobody intended. The alert draws on misconduct patterns from internal reviews over the past two decades and lists areas firms may want to examine. Translation: supervision duties do not pause for new tools — supervision is the reason the tool is allowed at all.

WHY IT MATTERS

Add AI use to your internal review checklist now. A control nobody tested is a finding waiting to be written.

7 Microsoft Gave Its Models a Rulebook That Beats Your Prompt

TechCrunch · Sep 14 · relevance 22/30

Microsoft published an AI code of conduct for its own models. It sets absolute constraints against cyberattacks, nuclear weapons and deepfake production. It also forbids models from using deception or collusion to escape human oversight. Each model gets an overarching code that overrides individual user preferences and any specific task. Microsoft CEO Satya Nadella welcomed 'embedded evaluators' as a way to make those promises testable. Translation: vendors are writing the rules your AI tools will follow — and you are not in the room.

WHY IT MATTERS

Ask your AI vendors for their model rules in writing. Constraints that never surface are constraints you cannot rely on in a dispute.

THE BOTTOM LINE
The Opportunity
→ Check every citation, quote, and witness name in an AI-assisted draft against the primary record before you file.
→ Ask whether AI-driven processes materially drive your reported numbers — and whether that role is disclosed.
→ Write down your supervision rule for employee AI use, and keep the review record.
The Risk of Waiting
Sanctions are now cheap to trigger and expensive to explain. In both cases this week, opposing counsel found the fabrication — not the lawyer who filed it. Courts are also asking what procedure existed before the mistake, so a policy written afterwards reads as damage control, not governance. And where AI drives reported results, the omission tends to become the claim. The cost compounds quietly: one unchecked draft, one unwritten rule, one renewal you did not test.
My Take
Everyone is arguing about which AI tool to buy. The tests that mattered this week were about verification: who reads the record, and who signs off. Morgan & Morgan's $1 billion bet assumes the firm owning the workflow can sell it. Eight made-up cases sit on its record, and the bet still makes sense. But the discipline that turns AI into an asset is boring and human: audit the cites, name the reviewer, keep the log.
✨ KEY TAKEAWAYS
1 New Mexico's top court fined a lawyer $5,000 for testimony that AI invented in a murder appeal.
2 A Los Angeles associate fined for fake citations filed a plan for verifying authorities.
3 Morgan & Morgan will spend $1 billion on AI and sell its platform to other firms.
4 94% of law firms use or explore generative AI; Microsoft Copilot leads at 76%.
5 This week's gap was not adoption but verification — who checks the record.

Forward this before your rival forwards it. 📩

One forward takes five seconds — and might save a colleague a sanctions hearing. Or send them to admissible.beehiiv.com. Three emails a week — the legal AI news that actually matters, in plain English.

💬 Hit reply with the one AI question no one at your firm will ask out loud — every reply helps us cover what matters most to you next time.

ADMISSIBLE AI
Curated from 13+ legal-tech sources, three times a week. Reply anytime — it genuinely shapes what we cover.
You're receiving this because you subscribed at admissible.beehiiv.com.

Keep reading

View more
caret-right
© 2026 Admissible AI.
beehiivPowered by beehiiv