Adams v. Kiewit Infrastructure (Arizona 2026): Court Warns Pro Se Plaintiff After AI Hallucinations — Next Violation Will Be Sanctioned | Advocate Prakhar

Case at a Glance

Background

Pro se plaintiff Bathsheba Nichole Adams filed a Motion for Leave to File Revised Second Amended Complaint in a pending employment and civil rights action against Kiewit Infrastructure West Company. The court reviewed the attached proposed complaint and identified two specific citation problems: a citation to “Ellsworth v. Am. Arbitration Ass’n, 148 Ariz. 17, 19 (App. 1985)” that referenced a non-existent Arizona case (the actual case at that citation was Ellsworth v. Am. Arb. Ass’n, 148 P.3d 983 (Utah 2006), a completely different case from a different jurisdiction involving arbitration); and a citation to “Aguilar v. ASARCO” at 780 F.3d 1101, a 9th Circuit citation that does not exist (the actual case at that citation was Lary v. Trinity Phys. Fin. & Ins. Services, 780 F.3d 1101 (11th Cir. 2015), a completely different circuit and party).

The court additionally noted that the motion failed to comply with Local Rule 15.1(a) (requiring a proposed amended pleading to show changes in track-changes format) and that there was no indication Adams had met and conferred with defendants as required before filing a motion to amend. The court struck the motion on these combined grounds.

In addressing the citation problems specifically, the court issued a forward-looking warning about the consequences of further AI-generated hallucinations, citing previous Arizona district court decisions on the same issue.

The AI Issue

The court stated directly that “these hallucinations allow the Court to infer that Plaintiff prepared her filing with the use of generative artificial intelligence.” This represents courts moving from suspicion to express inference: the dual-hallucination pattern (a case existing but in the wrong jurisdiction, and a citation that exists with a completely different case) is recognized as a generative AI fingerprint. The court then framed Rule 11’s requirements: the duty to verify applies to AI users and non-AI users alike, and pro se status is not a shield. Future hallucinated citations will result in sanctions.

What the Court Decided

• Two AI-hallucinated citations in a single motion — one citing a real case from the wrong jurisdiction, one fabricating a non-existent case — are sufficient for the court to infer AI use without further investigation.
• Rule 11 requires a party or attorney to sign every filing with a certificate that includes verification of all citations for accuracy; AI-generated citations that have not been independently verified violate this standard.
• Pro se status does not relieve a party of the Rule 11 verification obligation — the same standards apply to all filers.
• The immediate sanction was striking the non-compliant motion from the record; the court issued an explicit prospective warning that future hallucinated citations will result in Rule 11 sanctions.
• The court cited Ghadimi v. Arizona Bank & Trust (2025) for the principle that filing fictitious cases “results in confusion and unnecessary work for opposing parties and the court.”

“These hallucinations allow the Court to infer that Plaintiff prepared her filing with the use of generative artificial intelligence. While the use of such a tool is not necessarily problematic, the failure to confirm the accuracy of the filing is.”

— Judge Humetewa, D. Arizona, Adams v. Kiewit Infrastructure West, April 29, 2026

The India Angle

Indian Law Equivalent

Under BCI Rules and Indian court practice, an advocate’s duty to verify citations is absolute. In B.D. Kaushik v. State of Uttarakhand (2012) 12 SCC 1, the Supreme Court reiterated that advocates owe a primary duty of honesty to the court. A pro se litigant in India who cites non-existent authority — even inadvertently — may face the court’s inherent jurisdiction to dismiss the petition or impose costs under Section 35A CPC for false and vexatious claims.

Bar Council Rules

BCI Rules, Chapter II, Rule 22 requires that facts stated in pleadings be true to the advocate’s best knowledge and belief. For self-represented litigants, the same principle applies through the court’s inherent contempt jurisdiction. Rule 11’s equivalent in Indian practice is the overarching duty not to mislead the court enforced directly by the presiding judge.

Practical Advice for Indian Advocates

• When using AI to draft pleadings in India, verify every case citation against SCC Online, Manupatra, or the court’s official judgment database before filing — wrong jurisdiction or wrong party citations are AI hallucination hallmarks.
• Be especially careful with citation format: Indian AI tools may generate plausible AIR, SCC, or SCR citations that do not correspond to any real judgment. Cross-check the parties’ names, the year, and the page number against the actual database entry.
• If a court strikes or rejects a filing due to citation errors, use the opportunity to conduct a full audit of all pending filings before the next hearing — proactive correction prevents escalation from a warning to sanctions.

Quick Takeaways

• Courts can now infer AI use from hallucination patterns without requiring the party to admit it.
• Pro se status provides no Rule 11 exemption — all filers must verify citations independently.
• A struck filing with a warning is the first step; next violation brings monetary sanctions.

Deep Dive: Inferring AI Use from Citation Patterns

Adams v. Kiewit is notable for the court’s explicit statement that hallucinated citations allow it to infer AI use. This shifts the AI hallucination analysis from a factual question (did the filer use AI?) to an evidentiary one (does the citation pattern indicate AI generation?). Courts are developing pattern recognition: a citation to a real case from the wrong jurisdiction (Ellsworth), paired with a citation to a non-existent case at a real reporter volume and page number occupied by a completely different case (Aguilar), are the kind of errors that human researchers generally do not make but AI tools consistently do.

This evidentiary shift has practical consequences. Once a court has inferred AI use from the citation pattern, subsequent filings by the same party are subject to heightened scrutiny. The court in Adams essentially put the plaintiff on notice that it is now watching for the same pattern in future filings, and that finding it again will result in sanctions rather than just a struck pleading. The move from inference to enforcement is rapid once the initial pattern is identified.

The broader significance of the ‘hallucination as inference’ standard is that it removes the need for an admission or confession. Courts do not need to ask ‘did you use AI?’ — they can identify the AI use from the content itself and proceed directly to the Rule 11 analysis. This simplifies the sanctions inquiry considerably: the court does not need to resolve disputed facts about the drafting process; it just needs to confirm that the cited authorities are inaccurate in the AI-characteristic way.

For Indian advocates using AI drafting tools, the Adams case is a reminder that courts are increasingly sophisticated about AI content patterns. Indian judges who review large volumes of petitions are beginning to notice the same telltale signs — plausible but nonexistent citations, real case names with wrong reporters, correct propositions with wrong citations. The best protection is not to avoid AI but to verify systematically and thereby ensure that nothing reaching the court bears the hallucination fingerprint.