Comment on Proposed Tenth Circuit Rule 46.5(B)(4)

Respectfully, the Tenth Circuit’s Proposed Rule 46.5(B)(4) skates to where the puck was two years ago.

I am a member of the District of Columbia Bar and the founder of Showalter PLLC, a law firm I launched in March 2026 and built around an AI-native workflow. I am also the co-founder of Showalter AI Consulting, which trains law firms and in-house teams on how to triple productivity like I have. I make the assessment below as the first appellate lawyer in the country to publicly build a new law firm on that model.

I support the Court’s objective of eliminating sloppy mistakes in appellate filings. I write because the Proposed Rule as drafted identifies the wrong culprit, will not survive its own first years, and will impose costs that fall hardest on the litigants who can least afford elite counsel. It is also unnecessary, because this Court’s existing Rule 46.5(B)(2) already requires every filer to verify citations.

AI presents the opportunity to fix the longstanding and grievous injustice that ordinary Americans often cannot obtain elite legal representation for their meritorious claims and defenses because prices are so obscenely high. Those ordinary Americans are the ones who would bear the cost of the Proposed Rule if promulgated.

I offer my testimony partly because it is the kind the Court is least likely to receive. Most lawyers who comment on rules governing AI do not use AI, or do not know how to use it well. A court proposing to regulate a technology should want to hear from the people who operate it daily, and it should weigh that testimony accordingly, because the alternative is writing a rule about a capability from the outside.

Part I sets out the flaws in the Proposed Rule. Part II offers two ways to fix subsection (B)(4).

I. THE PROBLEMS WITH THE PROPOSED RULE AS DRAFTED.

A. In months, AI cite-checking tools will outperform even the most scrupulous human paralegal.

When it comes to citation mistakes, AI is not the problem but the solution. Cite-checking software is not a coming development. I already use AI cite-checkers every week on real filings, and they are already very close to bulletproof. There isn’t generally much for them to do—my Claude has not fabricated a case or quotation in many months. But if it did, my cite-checking skill would catch it.

I know this because this summer my Claude skill flagged in three minutes that my Ninth Circuit opposing counsel rested a central argument in its answering brief on a hallucinated quotation. I highlighted the hallucination in my reply brief, and opposing counsel moved for leave to file an amended brief.

We’re at an inflection point. In 2024, AI allowed sloppy lawyers to shotgun their sloppiness by producing more unchecked assertions than they ever could before. In 2026, however, AI is catching those sloppy lawyers’ mistakes. In 2028, the most scrupulous human paralegal in the world will be a far less reliable cite-checker than AI.

Today, the only mistakes remaining after my AI cite check are the ordinary mistakes present in nearly every brief ever filed—an italicized word that should be roman, a pincite off by a page, a Bluebook abbreviation missed. I find fewer of these mistakes after my AI cite-check than I did with the human paralegals I worked with in DC BigLaw. AI citators are already reducing these ordinary citation mistakes below the pre-AI level and soon will eliminate them altogether.

The Court need not take my word for it. Whether a verification tool detects a given defect is a measurable fact. Take a brief whose errors are already known, run the tool, and count what it returns. What I describe from daily use is an observable property of software that any member of this Court can confirm in an afternoon, and I would be glad to demonstrate it.

One tool I use, Midpage, is an AI legal research and drafting platform that will cite-check a draft brief, and it hyperlinks its responses to the cases on which they rely so that a lawyer can confirm each one. Its AI-powered citator also flags cases that later courts have treated negatively.¹ Another tool, BriefCatch, is a widely used brief-editing program that released a citation-verification tool called RealityCheck in March 2026. BriefCatch reports that, applied to a brief filed in Fletcher v. Experian Information Solutions, Inc., RealityCheck identified every legal-authority error later cited by the Fifth Circuit, and also found seven additional errors the court did not mention.² That occurred in March 2026. And as the technologists often say, these tools are the worst we’ll ever use.

As I’ve been told by the vendors, these citators work in two layers. The first is deterministic, which means it involves no probability and no generation: the reporter volume, court, and case name in the brief are compared against the entries in legal databases, and the comparison either matches or it does not. The same input returns the same output every time. A deterministic check cannot hallucinate, because it is not producing text; it is looking up a record. That layer disposes of the fabricated citation. The second layer is analytical and does use a model. It addresses the question whether the authority supports the proposition, which requires reasoning.

A rule premised on the human eye being the only trustworthy check is therefore already behind the tools available to any lawyer who cares to use them.

B. The Proposed Rule misidentifies the culprit.

By its terms, the Proposed Rule addresses only documents “prepared with the assistance of generative artificial intelligence.”³ Filings prepared the dated way apparently do not face the Proposed Rule’s constraints. But sloppy lawyering did not suddenly enter the world in 2022. The Court’s response should target sloppiness, not AI.

Two years before ChatGPT was released, Professor Douglas E. Abrams catalogued dozens of federal and state court opinions from the 2010s calling out specific lawyers for sloppiness and poor advocacy. Courts described lawyers’ written work as “slipshod,” “shoddy,” “sloppy,” “careless,” “strikingly inadequate,” “worthless,” “lax draftsmanship,” “marred by unprofessional errors,” showing “lack of effort,” “glaring errors,” “rampant deficiencies,” and “somewhat appalling.” One court reported a lawyer’s failure to “read the record closely or proofread his own briefs” or to meet “minimal standards of grammar.”⁴ Across the approximately 50 cases Abrams cited, as best I can tell, one lawyer was sanctioned $200—the rest faced no court sanction or bar discipline.

Some judges and lawyers have stigmatized AI use by focusing on misuse and treating AI-related errors more harshly than comparable human-only errors. One lawyer recently pleaded with a judge that her mistake came from her own pen, as if that made it better.⁵ That plea underscores that our legal culture has begun to treat an error as more grievous when it originated with AI. I believe that these double standards are rooted at least partly in a fear of change. That fear is understandable: whether we like it or not, AI will profoundly reshape the legal profession in the near future.

We should resist double standards that stigmatize AI use anyway. There is a simple solution to hallucinations and other misstatements: remove them. And that is true wherever the error came from. The ad hominem fallacy is when a speaker avoids an argument’s substance by attacking its human source; I use the term “ad machinam fallacy” for the same mistake when the source is a machine. An argument’s soundness is determined by the strength of its reasoning, not where it came from. The same logic applies to errors and misstatements—what matters is not the error’s source but whether the lawyer exercised the diligence to identify and eliminate it.

C. The Proposed Rule duplicates duties this Court already enforces.

Existing subsection (B)(2) already imposes the verification duty that proposed subsection (B)(4) would add. In Lichfield v. Kubler, the Court explained that under existing subsection (B)(2), “a bare-minimum reasonable inquiry will ensure legal authorities cited in papers submitted to the court actually exist, and the quotations or propositions attributed to them also exist.”⁶ The Court added that “[c]ounsel is responsible for ensuring briefs and representations made to the court are warranted by the law and supported by the record, no matter the technology used to create them.”⁷ The Court has enforced that duty with the tools it already has, sanctioning a lawyer who cited nonexistent cases while observing that “there is nothing inherently problematic with the use of GenAI in the practice of law.”⁸

The Fifth Circuit considered a rule very similar to the Proposed Rule and declined to adopt it. Its proposal would have required counsel and pro se litigants to certify either that no generative-artificial-intelligence program was used or that “a human checked the AI-generated text for accuracy.”⁹ After public comment, that court “concluded that existing rules were sufficient to deter misconduct related to generative AI use, without the need for a rule specific to generative AI.”¹⁰ When a lawyer later filed a brief with fabricated quotations, the court sanctioned her under existing authority, explaining that “the same sanctions rules apply, and the rules we have are well equipped to handle these types of cases.”¹¹

D. The Proposed Rule taxes access to justice.

If we misidentify AI as the cause of sloppy lawyering, we could suppress tools that will enable us to crush litigation expense and aid deserving Americans—meritorious plaintiffs and meritorious defendants both—who would otherwise be forced to acquiesce to unjust outcomes for the sake of their pocketbook.

The Proposed Rule taxes access to justice. I founded Showalter PLLC to pass along to clients the cost savings AI makes possible. I have charged $12,500 for a federal court of appeals opening brief that a traditional elite firm likely would have priced at $250,000 to $400,000. I charged $20,000 for a Supreme Court amicus brief that typically would have cost around $200,000. I charged $5,000 for an answer that typically would have cost $30,000 or more. Whatever effect AI has on my earning potential in the long run, it is awesome that I am now able to provide legal services for Americans who in the past would never have been able to afford elite work.

We should lean into that, but the Proposed Rule leans against it. Manual cite-checking is a highly mechanical yet still expensive component of brief production. A rule that compels a human to repeat, by eye, work a tool performs at least as accurately does not improve any filings but raises the price of every brief.

The biggest companies and richest individuals probably will not notice the cost of a paralegal charging a few hundred dollars per hour. Ordinary middle-class Americans do.

II. RECOMMENDED CHANGES.

I propose three changes to the proposed text. First, I propose dropping the technology trigger, so that the certification applies to every filing and does not contain double standards or stigmatize AI use. Second, I propose certifying results. The Proposed Rule requires certifying that “a human” performed a review. My proposal would require certifying that the work is right, which is all that matters. Third, I propose omitting verification of “legal analysis” and of “compliance with all relevant rules and ethical obligations.” Subsection (B)(2) already requires that the issues presented be “warranted by existing law,” and every lawyer who practices before this Court is already subject to discipline for failing to comply with its rules. Restating those duties can only create confusion.¹²

Alternatively, I propose that the rule endorse verification through AI tools that do not ever miss a fabricated case or quotation. If an AI tool doesn’t miss, there is no plausible reason for a human to check its work. If an AI tool does occasionally miss, the lawyer cannot rely on it to satisfy the rule.

I would welcome the opportunity to provide additional detail, including demonstrations of the verification tools described above, if that would assist the Court.

Respectfully submitted, Michael J. Showalter Showalter PLLC Showalter AI Consulting