The Texas-Level Shift Law Firms Cannot Ignore Right Now

What Changed In California And Why Texas Firms Should Care

What changed in California is not merely the existence of AI guidance. What changed is that the discipline story moved from policy discussion into enforcement. In April 2026, California Courts Newsroom reported that the State Bar had taken action over false or AI-generated citations, and that the broader conversation had shifted from “what if AI gets it wrong?” to “what happens when a lawyer files it anyway?” [source] [source].

The reason Texas firms should care is practical. Dallas-area firms regularly litigate in state court, federal court, and appellate venues where a single flawed citation can draw immediate scrutiny. A fabricated case name or a misquoted proposition is not just a drafting error; it can undercut the credibility of the lawyer, the client’s position, and the entire filing team. If the mistake lands in a court record, the repair cost rises fast. That can mean corrective filings, awkward client conversations, motion practice from the other side, and a question from a judge or disciplinary authority about why the citation was not checked before filing.

What this changes operationally

California’s own rulemaking activity reinforces the same point. The State Bar had a proposed AI rules package out for public comment, showing that regulators are thinking beyond isolated disciplinary matters and toward broader governance of generative AI in legal work [source]. In other words, this is not a one-off story about one bad brief. It is a sign that professional oversight is catching up to the speed of AI-assisted drafting. Texas firms should read that signal as a reason to build controls now, not after the first complaint.

There is also a local relevance point for Dallas and other Texas metros. Firms that handle federal litigation should pay close attention to Northern District of Texas practice, because that court requires disclosure when a brief is prepared using generative AI [source]. That local rule does not eliminate the underlying ethics risk. It simply shows that courts are beginning to formalize expectations. For firms serving clients in Dallas County and beyond, the compliance question is no longer whether AI will be reviewed. It will be reviewed. The question is whether your workflow can survive that review.

Why AI Hallucinated Citations Are A Professional Responsibility Problem

An AI hallucinated citation is exactly what it sounds like: the model produces a case, quote, reporter reference, statute, or proposition that looks plausible but is wrong, nonexistent, irrelevant, or lifted out of context. The danger is that legal writing rewards confidence in appearance. A fake citation can look polished enough to pass a quick glance, especially when it is embedded inside a larger, otherwise competent draft. That is why the risk is not limited to beginners or solo practitioners. Even experienced litigators can miss a fabricated authority if they trust the model too much or rush the review.

From a professional responsibility perspective, the problem touches competence, supervision, candor, confidentiality, and billing. Texas Ethics Opinion 705 says lawyers must have a reasonable and current understanding of generative AI, cannot blindly rely on its outputs, and remain responsible for the work product they submit [source]. California’s proposed AI comments likewise say lawyers must “independently review, verify, and exercise professional judgment” over AI output used in representation [source].

What to do next

“must independently review, verify, and exercise professional judgment”

That language matters because it captures the core issue. AI can assist with drafting, brainstorming, and summarizing, but it cannot own the legal judgment. The lawyer must do that. California’s practical guidance also warns that generative AI outputs can be false, inaccurate, or biased, and that those outputs need to be critically analyzed, supplemented, and improved when necessary [source]. In plain English: if the citation looks right but has not been independently checked, it is not ready for a filing.

The consequences of getting this wrong extend beyond the court. Clients may question whether the firm is actually doing the work it billed for. Opposing counsel may use the mistake to attack credibility. Judges may start reading the rest of the brief more skeptically. And if the false citation is tied to a legal proposition that matters to the outcome, the harm can become substantive, not just reputational. That is why hallucinated citations belong in the same risk bucket as other preventable filing errors: they are workflow failures, not just software glitches.

There is also a confidentiality dimension. The California practical guidance cautions lawyers not to input confidential client information into a generative AI system without adequate security and confidentiality protections [source]. That means the same tool that might fabricate a citation can also create a data-handling issue if a lawyer pastes in sensitive facts, strategy, or protected documents without vetting the platform first. So the professional responsibility problem is broader than legal accuracy. It is about the whole chain of judgment.

A Practical AI Citation-Check Workflow For Texas Law Firms

If the discipline warning tells Texas firms anything useful, it is that AI should never be the last step in the drafting process. It should be the first pass. A defensible workflow starts with a simple principle: AI may help generate text, but only a human lawyer can decide whether the text is accurate enough to file. That means every legal authority in an AI-assisted draft should be traced back to a primary source before the document leaves the firm.

A workable workflow for a Dallas or Texas litigation team might look like this. First, the lawyer or paralegal uses AI only for brainstorming, issue spotting, summarization, or a rough draft. Second, the drafter creates a source list from primary materials: cases, statutes, rules, and docket materials. Third, an associate or paralegal checks each citation against the original source, including case name, reporter, court, date, page, pinpoint citation, and the proposition for which it is cited. Fourth, the supervising attorney performs a final read focused specifically on accuracy, tone, and legal support. Fifth, the filing attorney signs off on the final version and retains a record of the verification steps.

What to do next

That process sounds basic because it is. But basic is the point. Texas Opinion 705 emphasizes that lawyers are responsible for submitted work product even when the original drafting or research came from technology [source]. The California guidance similarly supports a critical review and verification process rather than passive reliance on AI output [source].

The workflow should also be documented. Keep the prompt, the AI output, the redlined edits, the source-check notes, and the final sign-off in the matter file or a secure internal repository. That record matters for three reasons. First, it helps the firm audit its own process. Second, it gives the firm a way to answer client questions about how the work was prepared. Third, it creates a defensible record if a court or insurer later asks what happened. The goal is not to create paperwork for its own sake. The goal is to prove that the firm used a human-controlled process, not an unattended machine pipeline.

One more point: do not let speed become the excuse. AI can save time on initial drafting, but the saved time is only real if the verification system is disciplined enough to preserve accuracy. In litigation, a fast but unverified draft is not a productivity gain. It is a liability waiting to happen.

Checklist Table For Safe AI Use Before Anything Gets Filed

The simplest way to turn this topic into day-to-day behavior is a short checklist that sits between drafting and filing. Partners, associates, and legal operations staff should be able to use it quickly, without reading a long policy manual. The table below turns the California warning and Texas guidance into a practical pre-filing control set.

There is one additional point Dallas-area federal litigators should not miss: the Northern District of Texas requires a disclosure on the first page of a brief prepared with generative AI under the heading “Use of Generative Artificial Intelligence” [source]. That does not just create a disclosure obligation. It creates a habit of thinking about AI use before filing. Firms that already have a strong checklist will be in better shape if they practice in that court or in any venue that later adopts similar rules.

California’s guidance also supports the same control structure. It says lawyers should critically review AI outputs, verify them, and exercise judgment rather than delegating that role to software [source]. A good checklist makes that expectation visible and repeatable. A bad checklist is a policy on a shelf that no one uses when the filing deadline is near.

Supervision, Training, And Approval Rules Managing Partners Should Set

Most AI failures in law firms are not really technology failures. They are supervision failures. That is why managing partners should put a named owner on the AI policy and require practice-group leaders to enforce it. California’s practical guidance recommends clear policies, training, and supervision measures for permissible generative AI use [source]. Texas Opinion 705, meanwhile, makes clear that competence with technology is a continuing obligation, not a one-time box to check [source].

A sound governance model usually starts with approved use cases. For example: internal brainstorming, issue outlining, first-pass summaries, and formatting assistance may be permitted. But raw AI text should never be filed without human review. If the work involves dispositive motions, appellate briefs, class actions, or any document where citation quality is outcome-critical, the approval standard should be higher. A partner or designated ethics attorney should be able to require extra review before the draft goes out.

Signals fleets look for

Training also needs to be role-specific. Associates should know how to spot fabricated citations and why “looks right” is not enough. Paralegals and legal assistants should understand where they can help and where legal judgment must stop. Contract lawyers and outside support staff should be told what tools are allowed and what data is off limits. One-and-done lunch-and-learn sessions are not enough. The guidance should be refreshed as tools, court rules, and client expectations change.

The firm should also define escalation rules. If a lawyer cannot verify a source quickly, the rule should not be “guess and move on.” It should be “pause and escalate.” That may mean asking a supervising lawyer, checking a subscription database, or pulling the source from the reporter or docket. The point is to remove ambiguity under deadline pressure. Most poor AI decisions happen when a team is rushed and no one feels empowered to stop the filing.

Finally, the firm should assign responsibility for audits. A quarterly sample review of AI-assisted filings can catch pattern problems before they become incidents. That review should look for citation accuracy, documentation quality, disclosure compliance, and whether the sign-off process actually happened. The best policies are the ones that leave a trail of evidence, not just a promise that everyone did their best.

Client Communications, Billing, And Insurance Questions To Resolve Now

AI discipline risk is not only a filing issue. It is also a client-management and business-operations issue. California’s proposed comments suggest that a lawyer must communicate enough information about technology use when it presents a significant risk or materially affects the representation so the client can make informed decisions [source]. California practical guidance similarly says lawyers should consider disclosing AI use to clients and review client instructions that limit AI use [source].

For Texas firms, the first question is whether client notice is needed in the firm’s standard engagement documents. Some clients may not care whether AI helped draft an internal outline. Others will care a lot if the firm used AI on a sensitive matter, a high-stakes filing, or confidential material. A clear disclosure framework avoids ad hoc promises and inconsistent partner practices. The State Bar of Texas AI Toolkit is useful here because it includes sample client-disclosure forms and checklist materials that can be adapted into a firm policy [source].

What to do next

Billing is another pressure point. Texas Opinion 705 says lawyers may bill for actual time spent using AI and reviewing outputs, but not for time “saved” by AI [source]. That means firms should think carefully about how they describe AI-assisted efficiency to clients. A client who sees a bill that suggests the firm charged for work it did not perform will ask hard questions. The cleaner approach is to document the actual labor: prompt refinement, source checking, redrafting, and final review.

Insurance and records matter too. Malpractice carriers may want to know whether the firm uses generative AI, how it checks citations, and whether confidential client data is ever sent to third-party tools. Cyber coverage may also come into play if a platform retains prompts or uploads. Firms should review policy language before there is a claim, not after. Prompt logs, draft histories, and verification notes should be retained according to a clear policy so the firm can respond if a client later asks how a filing was assembled. In this environment, “we think someone checked it” is not a business control.

What Texas Law Firms Should Watch Next From California And The Texas Bar

The next 30 to 90 days matter because this issue is still moving. California’s AI rule package was in public comment, which means the exact final language could still shift [source]. That is important for Texas firms not because California text is binding, but because it may become a template that other bars and courts study. If the rule language hardens around independent review, verification, and supervision, those ideas are likely to spread.

Texas lawyers also have their own local signals to watch. The State Bar of Texas held an “Ask a Human” AI webinar on April 15, 2026, which shows that Texas ethics and practice leaders are actively engaging the issue [source]. That is a reminder that firms do not need to wait for a crisis to start building policy. They can use existing bar resources now and update them as new guidance appears.

What to do next

Dallas-area litigators should keep a special eye on federal brief filing rules in the Northern District of Texas because that court already ties AI use to disclosure in the filing itself [source]. Even if a firm mostly handles state-court matters, its federal practice can create a compliance standard that should be mirrored internally. If the same team files in multiple forums, the safest route is to adopt one uniform workflow that satisfies the strictest relevant requirement.

The near-term action plan is straightforward. Audit your current AI use. Identify where legal research is being assisted by generative tools. Verify whether every filing has a named human sign-off. Check whether the firm can prove source validation if challenged. Review client disclosures and billing language. And decide whether your current policy would still make sense if a judge, client, or bar investigator asked to see the process tomorrow. The firms that answer that question confidently will be the ones that turn this headline into a workable control system rather than a preventable ethics problem.

Frequently Asked Questions

Does using AI in a brief create an ethics problem by itself? No. The issue is not the tool alone. The problem arises when a lawyer does not verify the output, protect confidentiality, or ensure the filing is accurate before submission. Texas Opinion 705 emphasizes verification and responsibility for the final work product, and the Northern District of Texas requires disclosure when a brief was prepared with generative AI [source] [source].

What counts as an AI hallucinated citation in a legal filing? It is a citation, quote, case, or authority that the AI invents, misstates, or places out of context, even if it looks plausible on the page. California reporting described disciplinary action tied to false or off-point AI-generated citations, which shows how serious the problem becomes once the filing is made [source].

Should Texas firms ban generative AI from litigation work entirely? Not necessarily. A better approach is to limit it to draft support, require human verification, and set clear supervision rules before anything is filed. California practical guidance and Texas Opinion 705 both support responsible use rather than blind prohibition [source] [source].

Who should verify AI-generated case citations before filing? A lawyer must verify them, and the supervising or filing attorney should own the final sign-off. The Texas ethics opinion makes clear that the lawyer remains responsible for submitted work product, while California guidance says professional judgment cannot be delegated to AI [source] [source].

What written policy should a Texas law firm adopt first? Start with a short AI-use policy that covers approved tools, confidentiality, citation verification, supervision, disclosure, billing, and record retention. The State Bar of Texas AI Toolkit and Texas Opinion 705 provide a useful starting framework for that policy [source] [source].

Sources

• AI Hallucinations Put Three California Lawyers In State Bar Crosshairs
• Two Attorneys Face Disciplinary Charges, Third Agrees to License Suspension for AI Misuse
• Practical Artificial Intelligence in the Practice of Law
• 2026 Public Comment | The State Bar of California
• Ethics Opinion Issued by the Professional Ethics Committee for the State Bar of Texas
• Civil Rules | Northern District of Texas | United States District Court