Look, as I see it, $145,000 in Q1 sanctions should have been the wake-up call.

It wasn't.

According to a database maintained by lawyer Damien Charlotin, courts have now commented on 1,348 cases of AI-generated hallucinations in legal filings.

915 of those came out of U.S. courts.

The pace went from roughly two cases a week in early 2025 to two or three a day this spring.

As Eugene Volokh put it from UCLA Law, "by and large, many courts have taken substantial steps to punish people."

The question every Managing Partner should be asking on Monday morning is not whether your firm uses AI.

It is whether your firm has a workflow that can prove it didn't hallucinate.

Three Q1 rulings rewrote what "AI sanctions" means.

On April 4, Judge Mark Clarke in the District of Oregon hit two attorneys with $110,000 in penalties, the largest AI hallucination sanction in U.S. history.

The brief contained 23 fabricated citations and 8 invented quotations.

The attorney deleted the errors quietly rather than correcting them on the record.

Two weeks later, the Nebraska Supreme Court issued the first indefinite license suspension in the country tied to AI misconduct.

The attorney's brief had 57 of 63 citations flagged as defective.

He initially denied using AI.

And the firm everyone is talking about?

Sullivan & Cromwell.

In April, a Sullivan & Cromwell partner wrote Chief Bankruptcy Judge Martin Glenn an emergency letter about hallucinated citations in a Chapter 15 brief.

900 attorneys. Comprehensive AI governance. Mandatory training.

It happened anyway.

Sullivan & Cromwell has a written AI policy.

Gordon Rees Scully Mansukhani has one too, and that AmLaw 100 firm has now been caught twice.

Both firms got hit anyway.

Alexandra Smyth, General Counsel at LexisNexis, named the failure point.

"The model cannot verify that the case it cited exists, that the case says what the brief claims, or that the case remains controlling authority," Smyth said.

"This is architectural, not a capability problem."

The policy lives in a binder.

The verification step depends on a tired associate remembering to do it at midnight before a deadline.

That dependency is the gap.

Policies define expectations.

Workflows enforce them.

Here's what we've seen work in firms that haven't ended up in the Charlotin database.

Five disciplines.

1. Citations get pulled from inside Westlaw, Lexis, or CourtListener.

Not the generative model.

The model can draft the argument.

The citation lookup happens in the actual research database.

2. A second AI is not a verifier.

The Cherry Hill attorney who used one AI to "verify" another got fined $5,000 in April.

3. One named verifier per brief.

A specific name on the cover memo, not "the team."

4. A 5-minute citation audit before any filing.

The audit log saves with the file.

5. Standing disclosure language in the firm's letter to courts.

Five disciplines. Nothing fancy. Most firms are missing three of them.

Pull the 1-page AI Verification Workflow we install at firms before their next filing.

SOP, audit-log template, and the disclosure language for your court letter.

Built to answer the question your malpractice carrier is starting to ask.

Malpractice carriers do not read industry newsletters.

They read sanction opinions.

Underwriters are starting to ask renewing firms whether they have a documented AI verification workflow.

The firms that can answer with a one-page SOP and an audit log keep their renewal terms.

The firms that say "we have a policy on our intranet" are about to get repriced.

This is no longer about the cost of using the wrong AI tool.

It's about the cost of having no documented workflow when the call comes.

Has your firm documented an AI verification workflow that an outside auditor could review tomorrow?

A) Yes - written SOP and an audit log
B) Yes - written SOP, no audit log yet
C) We have a "policy" but no workflow
D) Nothing in writing yet

[Reply with your letter choice] - I'll share the results in the next edition.

The Sullivan & Cromwell letter is the moment this stopped being a story about rogue solos.

Sanctions are a lagging indicator.

Bar associations and malpractice carriers are next.

The partners who win the next 12 months aren't the ones who block AI or deploy it fastest.

They're the ones who can hand a regulator a one-page SOP that proves how verification works.

Liam Barnes

If your firm is in the bucket running AI on legal work without a verification workflow, that's exactly what we install at Cyberaktive.

Grab some time to chat

(if you don’t see a suitable time, just shoot me an email [email protected])

How Did We Do?

Your feedback shapes what comes next.
Let us know if this edition hit the mark or missed.

Too vague? Too detailed? Too long? Too Short? Too pink?

👍 Loved it | 👎 Not for me | 💡 I Have a Suggestion

OR: [email protected]

Was this week’s newsletter forwarded to you?

Sign up, it’s free.