M&T Bank Stadium Urination Shows AI Hallucination Risk for MD Law

This Risk Digest analysis is not legal advice, and it does not assume that any charge will be filed from the reported M&T Bank Stadium incident. The useful legal question starts narrower: if a Maryland lawyer saw the July 20–21, 2026 reports of a man allegedly urinating on a family at a Morgan Wallen concert, including an 8-year-old girl, and then typed “m&t bank stadium urination charges public indecency” into an AI legal research tool, what could go wrong before anyone caught it? Public reports also said MDTA Police told the victim they could not arrest the person because officers had not personally witnessed the act; as of those reports, the matter was described as under investigation, with no established suspect identification or filed charges treated here as fact.[1]

That is an ugly but ordinary criminal-docket problem: reported conduct, a child present, a possible public indecency or indecent exposure theory, and a lawyer under time pressure. It is also the kind of query that invites overconfident AI output because it looks simple. A human reader may see “public urination” and reach for a quick misdemeanor answer. Maryland law does not let the analysis stay that loose.

The Maryland Statute Is the First Trap

The starting point is Maryland Criminal Law Code §11-107, not a generic summary of “public indecency.” Section 11-107(b) states that a person convicted of indecent exposure is subject to imprisonment not exceeding 3 years, a fine not exceeding $1,000, or both.[2] That is the ordinary statutory penalty a Maryland practitioner would check first.

The reported stadium facts, however, would make a careful lawyer keep reading. Section 11-107(c), added through HB5 with an October 1, 2024 effective date, creates an enhanced penalty when indecent exposure is committed with prurient intent in the presence of a minor who is at least 2 years old and more than 4 years younger than the defendant. The penalty there is imprisonment not exceeding 5 years, a fine not exceeding $10,000, or both.[2]

That last distinction is not cosmetic. The M&T Bank Stadium reports may suggest why §11-107(c) would be researched, but they do not establish the facts required to charge or prove the enhanced offense. A lawyer advising a client, screening a charge, or drafting a motion would need the statute, the charging documents if any, the evidence, and the procedural record. An AI answer that collapses those steps into “public urination in front of a child equals enhanced indecent exposure” would be doing more than summarizing; it would be silently deciding contested elements.

Why This Query Is Harder Than It Looks

The dangerous version of the query is not exotic. It might ask: “Under Maryland law, what charges and penalties apply if a man urinates on a family at M&T Bank Stadium and an 8-year-old is present?” That query asks the tool to combine a state criminal statute, a recent amendment, possible aggravating elements, and a live reported fact pattern. Each part increases the need for source-order discipline.

A competent answer has to anchor itself in the Maryland General Assembly text, distinguish subsection (b) from subsection (c), preserve the October 1, 2024 amendment date, and avoid pretending that the stadium reports amount to a filed prosecution. It also has to avoid borrowing from another state’s “public indecency” scheme, because that phrase is not a safe substitute for Maryland’s statutory language.

This is where AI legal research risk stops being abstract. A hallucinated case about “public urination before a child” would feel useful precisely because the facts are emotionally vivid and the statute has a newly important enhanced-penalty branch. The more convenient the answer appears, the more likely it is to survive into a memo, email, plea discussion, or filing unless someone checks every cited authority against primary law.

The Benchmark Numbers Do Not Need Exaggeration

The best-known benchmark figures are already serious enough. In testing reported by Stanford RegLab and Stanford HAI, Lexis+ AI hallucinated on 17% of queries, while Westlaw AI-Assisted Research hallucinated on 34% of queries.[3] Those figures should not be treated as a live July 2026 product audit; the testing context was May 2024, and legal AI tools have continued to change. But the study remains important because it measured legal research products that lawyers could plausibly trust more than a general chatbot.

The more tailored warning comes from the AI Law Librarians synthesis published in February 2026. It identifies jurisdictional complexity as one of the recurring hallucination patterns in legal research and says hallucination rates spike for state and local law, reaching 100% for some local statutes.[4] That does not prove that a query about Maryland Code §11-107 will hallucinate at any particular rate. It does place this query in the category where legal AI systems have been documented to perform worse: jurisdiction-specific law, often outside the cleanest federal-case retrieval lanes.

The M&T Bank Stadium fact pattern adds two complications that a benchmark chart cannot show by itself. First, the statutory amendment is recent enough that stale summaries may miss or blur the HB5 enhancement. Second, the facts invite the tool to infer legal significance from the presence of a child, even though the enhanced penalty turns on statutory conditions and prurient intent, not public disgust alone.

The Professional Risk Is Not Just a Bad Answer

The consequence changes once an AI answer becomes a lawyer’s citation. In October 2025, secondary reporting and professional commentary described the Appellate Court of Maryland’s decision in Mezu v. Mezu, No. 361, September Term 2025, as identifying citation irregularities in 11 of 27 cited cases, including entirely fabricated cases, and referring counsel to the Attorney Grievance Commission.[5] Because the available account here comes through MSBA, The Baltimore Banner, and Eccleston & Wolf rather than a directly reviewed slip opinion, that attribution matters.

Mezu is the Maryland-specific sanction pathway that makes this more than a technology-quality complaint. If a lawyer files a memorandum citing nonexistent Maryland cases about §11-107, prurient intent, minors, or HB5, the issue is no longer whether the AI tool was impressive in a demo. The issue is whether counsel satisfied the duties of competence, candor, and supervision before putting authorities in front of a court.

Maryland Rules 19-301.1, 19-303.1, and 19-305.3(b) supply that duty structure: competent representation, candor toward the tribunal, and supervisory responsibility for nonlawyer assistance. ABA Formal Opinion 512 likewise requires lawyers using generative AI to verify legal authorities, and MSBA AI Task Force guidance frames AI use around lawyer responsibility rather than delegation of judgment.[6]

The National Center for State Courts has also warned legal practitioners about AI hallucinations, reinforcing the institutional point: legal hallucination is not merely an accuracy problem inside a research interface. It becomes a court-system problem when invented or unreliable authority enters a filing.[7]

The Boundary Is Verification, Not Trust

For this particular query, the AI system cannot be allowed to finish the research. It may help surface issues, but it cannot be the authority for the statutory penalty, the amendment history, the meaning of prurient intent, or the existence of Maryland cases applying §11-107(c). Those have to be checked against primary sources and then against reliable case-law databases, with the jurisdiction and date filters visible to the lawyer doing the work.

• Do not cite a Maryland case unless the case exists in a reliable database and the quoted proposition appears in the opinion.
• Do not treat §11-107(b) and §11-107(c) as interchangeable penalty provisions.
• Do not treat the reported presence of a minor as proof of the enhanced offense.
• Do not let an AI tool convert an active investigation into a charged case.
• Do not assume current Maryland law from an answer that fails to show the HB5 effective date.

That is the minimum boundary between research assistance and professional exposure. The stadium incident is not an AI incident. It is not evidence that any Maryland lawyer used an AI product. It is not a confirmed prosecution. Its value is as a live stress test for a query type that combines the documented risk factors: state criminal law, recent amendment, fact-specific aggravation, and intense pressure to produce a quick answer.

A Maryland attorney cannot treat AI output on §11-107 and HB5 as research-complete. Mezu shows why the endpoint matters: a fabricated citation in Maryland appellate practice can become a referral to the Attorney Grievance Commission, not just an embarrassing correction.

References

1. Fox Baltimore, CBS Baltimore, USA Today, The Baltimore Sun, and Hoodline reports on the July 20–21, 2026 M&T Bank Stadium urination allegation.
2. Maryland Code, Criminal Law §11-107, Maryland General Assembly, https://mgaleg.maryland.gov/mgawebsite/Laws/StatuteText?article=gcr&section=11-107
3. AI on Trial: Legal Models Hallucinate in 1 out of 6 (or More) Benchmarking Queries, Stanford HAI, https://hai.stanford.edu/news/ai-trial-legal-models-hallucinate-1-out-6-or-more-benchmarking-queries
4. What the Science Says About Hallucinations in Legal Research, AI Law Librarians, February 19, 2026, https://www.ailawlibrarians.com/2026/02/19/what-the-science-says-about-hallucinations-in-legal-research/
5. Mezu v. Mezu coverage and analysis, MSBA, The Baltimore Banner, and Eccleston & Wolf, October 2025.
6. Overview of Ethical Considerations, MSBA AI Task Force; ABA Formal Opinion 512, American Bar Association, 2024.
7. Legal Practitioner’s Guide to AI Hallucinations, National Center for State Courts.