Texas Attorney General opinion KP-0536 reports citations it could not find in a request letter and a city brief

Key takeaways

• The opinion says the request letter and the City of Harlingen's brief cite cases that "do not appear to exist", and gives two examples.
• Citing disputed facts, documents it does not construe and missing organizational information, the AG says it cannot address the questions comprehensively. The phantom citations appear in a separate footnote; any causal role is our inference.
• The AG quotes a federal court that citing nonexistent case law is "a hallmark of generative AI use". It does not say AI was used here.
• It cites Texas Ethics Opinion 705 on verifying generative AI output under Rules 1.01, 3.01, 3.03 and 3.04.

What does Opinion KP-0536 say about the citations it could not find?

Footnote 4 of the opinion (pages 2 to 3) says that "you and the City support several key arguments and legal propositions with citations to case authorities that do not appear to exist." It gives two examples. The first is City of Galveston v. Firefighters' Loc. Union No. 571, 233 S.W.3d 527, cited in the request letter for the proposition that a civil service position may not be eliminated only to remove the incumbent or to contract out a historical function. The AG says that "the decision to which you refer is not available at the citation you provide, nor did our search for the decision yield any results."

The second is City of San Antonio v. Guerra, 976 S.W.2d 814, which the City of Harlingen's brief cites on the limits of municipal officials outside a department's chain of command. The AG says the same is true for that decision. It adds that these are "only a small sample of the phantom citations we encountered in our review of the request letter and briefing in this matter." The opinion does not list the others, so the full count is not stated.

Why does the opinion cite Holmes and Texas Ethics Opinion 705?

Right after the examples, the footnote says: "Citation to nonexistent case law is 'a hallmark of generative AI use.'" It attributes that language to Holmes v. Univ. of Tex. at Austin, No. 1:24-CV-1135-RP, 2026 WL 297630, at *5 (W.D. Tex. Feb. 4, 2026). We did not open that decision and report only how the opinion uses it.

The footnote then says that, as the Texas Committee on Professional Ethics explained in a recent advisory opinion, lawyers have an ethical obligation to independently verify generative AI outputs in their work product. Failure to do so, it says, could run afoul of the Texas Disciplinary Rules of Professional Conduct, "including but not limited to" Rules 1.01 (Competent and Diligent Representation), 3.01 (Meritorious Claims and Contentions), 3.03 (Candor Toward the Tribunal) and 3.04 (Fairness in Adjudicatory Proceedings). The cited source is Tex. Comm. on Prof'l Ethics, Op. 705, 88 Tex. B.J. 310, 312 (2025). Our article on Opinion 705 covers that instrument on its own terms.

The opinion does not state that AI produced these citations. It says the citations could not be found, and separately that such citations are a hallmark of AI use. A reader should keep those two statements apart.

How does the citation footnote relate to the limits on the answer?

The opinion does not say the missing citations caused its limited answer. It gives other stated reasons: it says the questions depend on disputed facts and on documents it does not construe, namely a collective bargaining agreement, a memorandum of understanding and city ordinances and charter provisions. It also says it lacked general information about the fire department's structure and that "this office cannot resolve disputed facts." The phantom-citation footnote (footnote 4) is attached to that passage and opens with "We also note", so it reads as an additional observation. Whether the missing citations contributed to the limited answer is our inference; the opinion does not say so.

The opinion then states: "we cannot address your questions in a comprehensive manner, nor can we affirm the validity of any particular action taken by the City or the VIA." It offers general guidance on the Fire Fighter and Police Officer Civil Service Act. Its summary says the Act authorizes a governing body to abolish classified firefighter positions in good faith, and does not give the Valley International Airport board or airport director authority to fill those positions. That civil service interpretation is unrelated to AI, and we do not analyze it further here.

What should lawyers and legislative staff do now?

The opinion creates no filing rule. It reports unverifiable authority in its record, so these steps follow from it as practical precautions rather than requirements.

• Before a request or brief goes to the Attorney General, open each cited decision in a primary or licensed database and confirm the reporter citation, court and year.
• Where an outside advocate or a city office supplied the citations, have the submitting lawyer re-check them instead of relying on the supplier.
• Treat any AI-assisted research as a lead, and record who verified each authority and when.
• Read Opinion 705 and Rules 1.01, 3.01, 3.03 and 3.04 against your office's AI use policy.

Related reading on how courts and bars treat the same problem: our Texas Fourth Court of Appeals article and our Fifth Circuit sanctions article.

What we did not verify

What we opened: the full text of Opinion No. KP-0536 as a PDF on the Texas Attorney General website, on 2026-10-25.

What we did not open: the request letter (its public URL is in footnote 1 of the opinion), the briefs (which the opinion says are on file with the Opinion Committee), the Holmes decision, Texas Ethics Opinion 705 itself, and the two named cases or any search results for them. We rely on the opinion's own statements that the decisions could not be found.

What we refuse to claim: that generative AI produced any citation, that any lawyer or office was sanctioned, referred or found in breach of a rule, how many citations were affected beyond the two named examples, or that the opinion binds anyone on the AI point.

Related briefings

• Texas Ethics Opinion 705 on generative AI
• Texas Fourth Court of Appeals considers striking a brief over hallucinated citations
• Fifth Circuit AI sanctions ruling in the Fletcher and Experian matter
• California bar rule on verifying AI output