Key Takeaways

• AI Hallucinations Are Your Legal Liability When AI fabricates citations or misquotes sources in a court filing, the court holds you — not the AI — responsible. Judges are issuing five- and six-figure sanctions, case dismissals, and disciplinary penalties with increasing frequency and zero tolerance.

• Honesty Is Non-Negotiable When Mistakes Happen The attorneys facing the harshest consequences aren’t just those who made errors — they’re the ones who denied, minimized, or obscured their AI use afterward. If a hallucinated filing slips through, complete candor with the court is your only path to damage control.

• No Firm Is Too Big, Too Experienced, or Too Sophisticated to Be Exposed From solo practitioners to Sullivan & Cromwell, AI misuse is an equal-opportunity liability. Every attorney must personally verify citations, check for judge-specific AI disclosure orders, and follow firm AI policies — every single time, on every single filing.

Here’s What Your Law Firm Should Do Right Now 

This isn’t surprising, and I must say, if I’m being honest, I’m experiencing a little schadenfreude watching this. As soon as AI tools became widely available to the general public, the number of pro-se lawsuits skyrocketed. Most of us had seen these types of complaints pre-AI, so imagine one of these bad boys on AI steroids. Frightening. On top of this, apparently, attorneys have felt free to jump into the melee and file with the court whatever slop their AI tools pump out. And the judges are pissed.

All across the country, we are seeing judges’ wrath against careless AI filings.

Recently, in Oregon, a US magistrate judge imposed a $96,000 fine in sanctions ($15,500 in disciplinary penalties plus $80,500 for OC’s attorneys’ fees) for 3 court filings containing 23 fabricated citations and 8 false quotations generated by AI, and the client’s case was dismissed with prejudice. In Alabama, a US district court judge sanctioned both an attorney and his law firm because neither the attorney nor the firm could demonstrate that its AI policies had been enforced or explain how it would prevent future violations after the attorney “intentionally misled the court” about his misuse of AI in filings and then “feigned contrition, obfuscated the truth, and changed his stories.” Sanctions totaled ~ $47,000.

The Three Key Takeaways and What You Should Do Next

The first is that all submissions, all citations submitted to the court, whether from AI or not, are considered the same as any other filing, with same truthfulness and candor required to the court. If your AI lies to the court, YOU lie to the court, and judges are getting increasingly angry about this.

The second is that if you do make a mistake and submit a hallucinated mistake, you must exercise complete candor with the court. The attorneys who are getting in real trouble are the ones denying AI usage or being cavalier about its usage.

Finally, while it may be a little soon to draft a full AI policy for your firm, you should implement a few guidelines for your attorneys immediately. Here are the guidelines our Chief Legal Officer has issued for our law firms.

Straight From the Pen of Our Chief Legal Officer

These reflect existing ongoing professional conduct obligations and US Legal Groups’ policies.

• Limited consumer AI for client matters. Consumer versions of any AI platform whose terms permit data use for model training or disclosure to third parties are not appropriate for input of client-specific fact patterns or information. General searches may be okay pursuant to the following bullet point. Examine the use terms of any AI platform closely and avoid use unless they guarantee no use of data.
• Verify every citation before filing. Every case citation, quotation, and statutory reference in any court filing must be personally verified by the signing attorney against the actual source. AI assistance does not transfer your Rule 11 or candor obligations. Remember that AI is not a proofreader of its own output, so asking one AI tool to verify the citations produced by another AI tool does not satisfy this obligation.
• Check judge-specific AI disclosure orders. At the opening of every matter, confirm whether the assigned judge has a standing AI disclosure order. This applies in all states.
• Advise clients at intake. Although Heppner involved a criminal case, the holding is based on universal privilege doctrines that may also apply to family law, estate planning, and probate. Clients should be explicitly advised not to use consumer AI to analyze their case, prepare for meetings with counsel, or process attorney communications. My concern here is a family law client who uses AI to draft a parenting plan or similar on their own initiative and then provides it to their attorney or an EP client, who puts specific names, distributions, etc., into AI, and that such prompts could become discoverable in the future.
• AI note-taking tools. These tools present some risk in their use. Several platforms currently marketed to law firms (such as AI meeting assistants) process and store call recordings on third-party servers under terms of service that may permit either disclosure and/or use as model training. Under Heppner’s reasoning, such a call may not be confidential and the transcript could be discoverable. We are currently in the process of evaluating either a compliant note-taking tool or building our own.

I Am a Believer in AI. That’s Exactly Why I’m Writing This.

I have been an early adopter of AI tools. I use them, I encourage others to explore them, and I believe they will continue to reshape the practice of law in meaningful ways. None of that has changed.

What I’ve always said alongside that, and what this situation makes painfully concrete, is that AI requires oversight. It is a capable assistant. It is not a supervising attorney. The moment you treat it as one, you’ve created exposure.

And lest you think this is a solo practitioner problem or a small-firm risk, consider what happened at Sullivan & Cromwell just last week. One of the most prestigious firms in the world, whose partners reportedly bill around $2,000 an hour in bankruptcy cases, submitted a court filing with more than 40 AI-generated errors, including fabricated citations. The partner responsible apologized in writing to a federal judge. The errors were caught by opposing counsel, not the firm. The firm’s own AI policies weren’t followed.

No firm is too sophisticated for this to happen. No attorney is too experienced. Be careful out there, and if you want to discuss how to implement AI into your firm further, we’d be glad to talk through it.