Verify or Get Sanctioned: What the Case Law Actually Requires

I've spent fifteen years in courtrooms, and I've never seen the law develop this quickly around a single issue.

Every month another sanctions opinion lands. Every month another attorney forwards it to me with some version of the same question:

"Are courts saying we can't use AI anymore?"

They're not.

They're saying something much simpler.

Don't sign what you haven't verified.

That's the doctrine emerging from nearly every sanctions opinion over the last three years. Courts are not punishing attorneys for using generative AI. They're punishing attorneys for filing briefs that contain fabricated authority, misquoted cases, or unsupported propositions because nobody took the time to verify them before signing.

That distinction matters. It is the difference between treating AI as a drafting tool and treating it as your attorney. One is entirely consistent with the Rules of Professional Conduct. The other has become one of the fastest-growing sources of sanctions in modern litigation.

The same tool that drafts your motion can also invent your citation.

That isn't a flaw unique to one platform. It's the nature of large language models. They predict language. They do not understand legal authority the way attorneys do. Which means the better these systems become at writing persuasive prose, the more dangerous they become when attorneys mistake confidence for accuracy.

That's the paradox.

A clumsy answer makes you skeptical. A polished answer earns your trust. After the hundredth correct citation, many attorneys stop checking the hundred-and-first. That's where the problems begin.

Strip away the technology and none of this is actually new.

• Rule 11 didn't change.
• The duty of candor didn't change.
• An attorney's obligation to conduct a reasonable inquiry before signing a filing didn't change.

The courts are simply applying longstanding principles to a new way of practicing law.

The Signature Is the Certification

One of the biggest misconceptions I hear is that sanctions begin when an attorney asks AI a question.

They don't.

The problem starts when the attorney signs the filing.

The Ninth Circuit said exactly that in LNU v. Blanche, where two attorneys were sanctioned, suspended from practicing before the court for six months, and ordered to pay monetary penalties after filing a brief containing fabricated authorities and misattributed quotations. The court did not criticize the attorneys for experimenting with generative AI during research or drafting. It focused on something much more familiar.

They signed the brief.

By doing so, they certified that the authorities they cited actually existed and supported the propositions for which they were offered. When that certification proved false, Rule 11 and the attorney's duty of candor became the issue, not artificial intelligence.

That's an important distinction because it reframes nearly every headline about AI sanctions over the past three years.

This is not a new body of law governing technology.

It's existing professional responsibility doctrine applied to a new source of error.

That is exactly how courts have approached these cases from the beginning.

Read Enough Opinions and Another Pattern Emerges

Read enough sanctions opinions and you stop seeing AI cases.

You start seeing lawyer cases.

The headlines talk about ChatGPT, Claude, or whatever model happened to be involved. The opinions almost never do. They spend remarkably little time discussing the technology itself. Instead, they focus on questions judges have been asking lawyers for decades.

• Did you read the case?
• Did you verify the quotation?
• Did you have a reasonable basis for signing this filing?

When the answer to those questions is no, the technology becomes almost irrelevant.

Something else becomes obvious, too.

Courts consistently punish what happens after the mistake more severely than the mistake itself.

That is the pattern hiding inside the pattern.

The hallucinated citation gets everyone's attention. The lawyer's response determines the sanction.

Some attorneys immediately investigate, notify the court, correct the filing, and accept responsibility. Others double down. They blame co-counsel. They blame staff. They blame the software. Some continue citing fabricated authority even after opposing counsel or the court has questioned it.

Those cases almost never end well.

The lesson isn't subtle.

Judges understand mistakes.

They have very little patience for concealment.

Candor Carries More Weight Than Perfection

Look at Dec v. Mullin, decided by the Seventh Circuit this past March.

Counsel cited two nonexistent authorities and included a fabricated quotation in the standard-of-review section of a brief. At oral argument, he initially denied using artificial intelligence. Later, he submitted a letter acknowledging exactly what had happened, accepted responsibility, and corrected the record.

The court admonished him and stopped there.

Why?

Not because fabricated citations are insignificant. They aren't.

Because the errors did not affect the outcome of the appeal, and more importantly, because counsel ultimately demonstrated candor instead of continued evasion.

Now compare that to Withers v. City of Aberdeen.

Four attorneys signed filings containing hallucinated authority. One attorney had already been caught making nearly the same mistake in another federal court. The problem wasn't simply that false citations appeared in the briefing. It was that the failure repeated itself after everyone involved knew exactly what had happened.

The result was dramatically different.

The court disqualified all four attorneys from the case and prohibited two of them from practicing before that district for two years.

Same technology.

Same category of error.

Completely different outcome.

Because courts weren't measuring the software.

They were measuring the lawyers.

The Cover-Up Is Often Worse Than the Hallucination

You can see the same principle running through nearly every major sanctions opinion.

In Twigg v. BSN Sports, Chief Judge Brann made clear that his frustration wasn't limited to fabricated authority. The opinion repeatedly returned to counsel's attempt to shift responsibility for the errors after they had been discovered.

In Flycatcher Corp. v. Affable Ave., counsel continued relying on defective citations after opposing counsel identified the problem and after the court specifically questioned the authorities. What began as inaccurate citations evolved into findings about supervision, professional responsibility, and public censure.

Go back to Mata v. Avianca, the case that introduced most lawyers to AI hallucinations in the first place.

People remember the fake cases.

Judge Castel's opinion spends just as much time discussing everything that happened afterward.

The court questioned the authorities.

Counsel defended them.

The court ordered copies.

Counsel produced fabricated opinions.

Only after repeated opportunities to correct the record did the full story emerge.

The hallucination created the problem.

The response created the sanctions.

That's why I tell lawyers to stop treating Mata as an AI case.

It isn't.

It's a candor case.

AI simply supplied the newest way to violate one of the oldest duties in the profession.

Verification Is Not a Checklist. It's the Workflow.

One thing about these opinions keeps getting overlooked.

Not a single court has said lawyers should stop using AI.

Quite the opposite.

Several opinions acknowledge that technology is now part of competent legal practice. Judges understand that lawyers use AI to brainstorm, summarize records, organize discovery, and generate first drafts. Those uses aren't controversial anymore.

What courts refuse to accept is the idea that AI can replace professional judgment.

That distinction matters because many lawyers still think verification is the last thing you do before filing a brief. Read the citations. Spot-check a few quotations. Make sure nothing looks obviously wrong.

That's not verification.

That's hope.

Real verification happens throughout the drafting process.

• Every authority should be opened and read before it's cited.
• Every quotation should be checked against the original opinion.
• Every proposition should actually stand for the point you're asking the court to accept.
• Every citation should come from a source you trust, not simply because a chatbot produced it with confidence.

That's how lawyers practiced before generative AI.

It's how we'll continue practicing after the current wave of sanctions opinions fades into history.

What Verification Looks Like in Practice

The courts have made one thing clear: verification cannot be an afterthought. It has to be part of the drafting process.

Here's how OurFirm.ai's Citation Verification works inside the litigation workflow, so every authority is checked as the brief comes together, not after it's filed.

The Better AI Gets, the More Discipline Lawyers Need

There's another paradox that deserves more attention.

The better these systems become, the easier it is to trust them.

Early versions of AI were almost helpful because their mistakes were obvious. They sounded robotic. Their legal analysis often fell apart after a paragraph or two. Most lawyers instinctively knew something wasn't right.

Today's models are different.

They write well.

They organize arguments well.

They often identify the right issues faster than a junior associate.

And that's precisely why verification has become more important, not less.

The greatest risk isn't that AI will produce terrible legal work.

It's that it will produce work that looks excellent until one fabricated citation, one altered quotation, or one misstated holding quietly makes its way into a signed filing.

The confidence is the trap.

The lawyer's signature is supposed to break that trap.

Strip Away the Technology

When I teach CLEs on AI, I usually end with the same thought.

Strip away the technology.

Forget the chatbot.

Forget the headlines.

Forget which model generated the text.

What's left?

A lawyer signed a document submitted to a court. That lawyer represented that the authorities existed, that the quotations were accurate, and that the law supported the arguments being made.

That responsibility has belonged to lawyers for generations. Artificial intelligence didn't change it.

It simply made it easier to forget.

The same tool that drafts your motion can invent your citation. That's why verification cannot be a feature bolted onto the end of the process. It has to be built into the workflow from the beginning.

I've spent fifteen years litigating cases before judges who expected lawyers to know their record, know their authorities, and stand behind every filing they signed. Nothing about generative AI changes those expectations. If anything, the recent sanctions opinions suggest they're becoming even more important.

• Read the case.
• Check the quotation.
• Verify the citation.
• Sign only what you know is true.

That's not a new rule for artificial intelligence.

It's competent litigation practice on a new medium.

See Verification the Way Courts Do

At OurFirm.ai, we built Citation Verification around the same principle the courts have been reinforcing in sanctions opinions across the country: trust, but verify.

The goal isn't to slow lawyers down. It's to make sure speed never comes at the expense of accuracy. AI should help litigators think faster, write better, and prepare more thoroughly. It should never leave them wondering whether the authority in their brief actually exists.

The strongest AI workflows don't replace professional judgment. They reinforce it.

Because when your name is on the signature block, verification isn't someone else's responsibility.

It's yours.