The Filing Needs a Receipt
The legal system is making AI-generated work carry an old institutional requirement: the signer owns the filing.

Lead image: legal evidence packets and verification paths for AI-assisted filing accountability.
The legal system is giving AI a very old answer: sign your name, carry the responsibility.
That is the useful signal from the current run of court orders, rule changes, and legal guidance around AI-generated legal work. The question is no longer whether lawyers, judges, clerks, or self-represented litigants will use AI. Many already are. The pressure is moving to the record: who checked the citations, who verified the facts, who read the authority, who caught the fake quote before it entered a public filing?
The Ninth Circuit’s June 3 order in LNU v. Blanche makes the line plain. The court sanctioned two attorneys after briefs included nonexistent cases, misattributed quotations, and misrepresentations of real cases. The panel said it was not punishing the simple use of generative AI. The violation happened at signing and filing. However the paper was prepared, the signature carried the lawyer’s responsibility for the accuracy of the filing.
That distinction matters beyond law. AI lowers the cost of producing institutional-looking text. It does not lower the cost of accountability. A brief can look polished while the citation layer is rotten. A memo can read confidently while the evidence chain is missing. A public system that cannot tell the difference has a real operating problem.
Florida moved that problem into a statewide rule. On May 28, the Florida Supreme Court amended Rule 2.515(d)(2), effective June 15, requiring the signer of any filing to represent that the legal authorities identified in the filing “exist and are accurately cited.” The rule applies to attorneys and unrepresented parties. It also authorizes sanctions after notice and an opportunity to be heard. The court framed the change as a uniform replacement for varied local AI-related orders.
The District of Kansas took a similar posture in Standing Order 26-01. The order warns lawyers and pro se litigants that AI-assisted filings still require review and verification of citations, quotations, paraphrased assertions, legal analysis, factual background, and procedural background. If the court believes that verification did not happen, it can strike filings, impose monetary sanctions, refer counsel to disciplinary authorities, disqualify counsel, impose filing restrictions, dismiss a case, or require sworn AI-use statements.
And the federal rules conversation is widening from citations to evidence itself. The Advisory Committee on Evidence Rules reported in May that it is still studying proposed Rule 707 for AI-produced evidence offered at trial without an expert, along with possible rule language for deepfakes. That is a different surface than fake case citations, but the operating question is close: when a machine output enters the courtroom, what foundation makes it reliable enough to matter?
The ABA’s courts update adds the institutional layer. It notes that judges are also experimenting with AI, that training remains uneven, and that AI-related proposals are moving through court rules, evidence rules, and congressional interest. A separate ABA Journal piece quotes Judge Scott Schlegel’s warning that when lawyers misuse AI, they may be sanctioned, but when judges misuse it, “it becomes the law.” That is the reason this belongs in Hypernovelty. The interface is moving faster than the proof culture.
Verification bottleneck
Verification is becoming the scarce institutional function.
- Legal text can be generated faster than courts, clerks, opposing counsel, clients, and self-represented litigants can verify citations, quotations, record references, and factual assertions.
- Court systems now have to verify two layers at once: the legal claim and the production path that created it.
- AI-produced evidence and possible deepfakes push the same problem into trial proof. The court has to know what the machine produced, how it was validated, and whether a human expert is needed.
- Watch next: whether courts turn AI verification into practical filing records, or leave each judge, clerk, lawyer, and pro se litigant to build their own ad hoc ritual.
Opportunities
Where value may appear: legal citation verification systems.
Someone could build tools or services that verify citations, attach source text, preserve filing provenance, flag hallucinated authority, check quotations against actual opinions, maintain reviewer logs, and create audit trails that clerks, lawyers, firms, legal aid groups, and self-represented litigants can inspect.
This is idea fodder only, not legal, compliance, procurement, financial, or investment advice. The practical operator test is simple: before a document asks an institution to act, show the receipt for the authority it relies on.
Sources
- Ninth Circuit, LNU v. Blanche, June 3, 2026: https://cdn.ca9.uscourts.gov/datastore/opinions/2026/06/03/24-4790.pdf
- Florida Supreme Court, Opinion SC2026-0673: https://flcourts-media.flcourts.gov/content/download/2489374/opinion/Opinion_SC2026-0673.pdf
- U.S. District Court, District of Kansas, Standing Order 26-01: https://ksd.uscourts.gov/sites/ksd/files/Standing%20Order%2026-01%20Use%20of%20AI.pdf
- U.S. Courts, Advisory Committee on Evidence Rules, May 2026: https://www.uscourts.gov/sites/default/files/document/advisory_committee_on_evidence_rules_may_2026.pdf
- ABA, “AI & the Courts: Recent Developments”: https://www.americanbar.org/advocacy/governmental_legislative_work/publications/washingtonletter/april-26-wl/ai-in-the-courts-update-0426wl/
- ABA Journal, “May It Please the Court: How do judges use generative AI?”: https://www.americanbar.org/groups/journal/articles/2026/how-do-judges-use-generative-ai/
- NYSBA, “Beyond the Mirage: Beware of Generative AI and Hallucinations”: https://www.nysba.org/beyond-the-mirage-beware-of-generative-ai-and-hallucinations/