The Sanctions Docket Is Growing: Why the Attorney Still Owns Every Citation in the Brief

One fabricated citation can now cost a lawyer five figures, a suspension, and a permanent line in a published opinion. That's the price of a single unverified paragraph a chatbot wrote at 11 p.m. before a filing deadline. The sanctions docket has stopped looking like a novelty and started looking like a trend line. Every new order reads the same way: the tool made it up, and the attorney signed it.

The rules haven't changed. What's changed is how easy it is to break them without noticing. Here's how the risk builds before the brief is written, during drafting, and after the filing lands in front of a judge.

The Duty Attaches Before You Open the Draft

Verification doesn't start when a citation gets pasted into a brief. It starts the moment an attorney decides to use an AI tool at all. Bar authorities have been unusually consistent on this point, and the guidance is not subtle.

As of early 2026, more than 35 state bars have issued formal guidance on generative AI in practice, and the through-line is the same across jurisdictions: understand the tool, verify the output, disclose the use when a court asks. That last point matters. A growing number of judges are asking, sometimes in standing orders that predate the brief you're about to file.

Before you prompt anything, three questions decide whether you're on the safe side of the ethics line:

• What the tool actually does. A general-purpose chatbot with no legal database behind it isn't a research tool. It's a text generator that will invent a citation before it will admit it doesn't know one.
• Where the data goes. If client facts leave the firm's environment, confidentiality is on the table before accuracy ever comes up.
• What the court expects. Standing orders on AI disclosure vary by judge and by district. Reading them is part of the filing, not an afterthought.

Hallucinations Sneak In During Drafting

The seductive part of AI drafting is that the output looks right. Case names read plausibly. Reporter volumes fall in the correct numeric range. Pin cites land on pages that could exist.

Then someone actually pulls the case, and half of it is fiction.

Consumer chatbots aren't the only offenders. A Stanford RegLab empirical study of purpose-built legal research tools found that even retrieval-augmented systems from major vendors hallucinated on a meaningful share of queries, enough that vendor marketing about "hallucination-free" output doesn't match what the products do in practice. The takeaway isn't that these tools are useless. The citation checker has to be the lawyer, every time, on every quote and every case.

Practical verification during drafting looks like this:

• Pull every case. Not the headnote the AI generated, the actual opinion in a real database. If it doesn't exist, delete it.
• Match every quote to a page. Fabricated quotations are the second-most-common failure mode after fabricated cases, and they travel just as far.
• Check the holding, not the summary. A real case cited for a proposition it doesn't stand for is still a Rule 11 problem.

The Sanctions Docket Keeps Growing After Filing

The consequences are no longer theoretical, and they're no longer isolated to a handful of viral cases. One recent Oregon matter produced a six-figure combined penalty against counsel after they filed fake citations and fabricated quotations across three summary judgment briefs.

Federal appellate courts are weighing in directly now, and the penalties have escalated beyond token fines. Discipline in recent orders has included monetary sanctions, referrals to state bar authorities, and multi-month suspensions from practice. Trade press reporting on attorneys sanctioned over reporting on attorneys sanctioned over AI use use has become a regular beat, and firms should assume any published sanctions order will surface in a client's diligence long after the fine is paid.

One recent wrinkle deserves attention on its own. In a September 2025 California appellate decision, the court sanctioned the filing attorney for two briefs full of fake citations, then declined to award fees to opposing counsel because they had failed to catch the fabrications themselves. Verification is starting to look like a duty that runs both ways.

The Attorney Owns the Signature, Full Stop

Signing a brief is an affirmation. The model doesn't sign anything, doesn't hold a bar card, and won't appear at the sanctions hearing. When the citation is wrong, the person who signed is the person who answers.

The firms handling AI well aren't the ones with the flashiest tools. They're the ones that have written down who verifies what, at which stage, before which filing, and then enforced it. That means human approval gates on any AI-assisted work product, private environments for anything client-adjacent, and audit trails a partner can hand a judge without flinching.

The technology can absorb the grunt work. It cannot absorb the responsibility.