The Attorney Who Doesn't Verify AI Is the New Attorney Who Doesn't Read the Cases

SavvyLex
Apr 13
6 min read

There is a version of this story that has become almost familiar.

An attorney uses an AI tool to research a brief. The tool produces confident, detailed citations — case names, reporters, page numbers, holdings. The attorney reviews the output for substance. The argument looks solid. The brief gets filed.

The citations are fabricated. The cases do not exist.

The court notices. The opposing counsel notices. The judge issues an order to show cause. The attorney explains that the AI tool produced the citations without warning. The court is not moved.

This has now happened enough times — in federal appellate courts, in state trial courts, in tax court, in family court — that it is no longer a cautionary tale. It is a pattern. And the legal profession's response to that pattern is becoming a formal obligation.

What ABA Formal Opinion 512 Actually Says

In July 2024, the American Bar Association issued Formal Opinion 512 on generative AI tools. It was the first formal ethics guidance the ABA had issued specifically addressing AI. The opinion is not long. Its core obligation is not complicated.

Lawyers must understand the capabilities and limitations of the AI tools they use. They must verify every output before relying on it. They must not submit AI-generated content to a court, a client, or any third party without independent confirmation of its accuracy.

This is not a new standard invented for AI. It is the application of an existing standard — ABA Model Rule 1.1, the duty of competence — to a new context.

Rule 1.1 requires lawyers to provide competent representation, which demands "the legal knowledge, skill, thoroughness and preparation reasonably necessary for the representation." Comment 8 to Rule 1.1, added in 2012, extends that duty explicitly to technology: lawyers must "keep abreast of changes in the law and its practice, including the benefits and risks associated with relevant technology."

The duty was always there. AI made it impossible to ignore.

The Sanctions Are Real and Accelerating

Law.com catalogued six separate AI sanctions cases in 2025 alone. The pattern is consistent across all of them: an attorney used a generative AI tool, the tool produced fabricated or inaccurate citations, the attorney submitted the work product without independent verification, and a court imposed sanctions.

In February 2026, a federal appeals court ordered a lawyer to pay $2,500 after submitting a brief containing AI-generated fictitious case citations — and expressed open frustration that this pattern continues despite years of high-profile warnings. The court noted that the Mata v. Avianca decision in 2023 should have made the verification obligation unmistakably clear to every practicing attorney.

In early 2026, the U.S. Tax Court found that an attorney had "most likely" relied on AI in citing nonexistent cases in support of a client's position. The court did not accept the ambiguity as a defense.

A California appellate court went further: it sanctioned lawyers for failing to detect fabricated citations in opposing counsel's brief. The verification obligation, in that court's view, runs not only to your own work product — it runs to what you accept, unchallenged, from the other side.

The Florida Bar issued Ethics Opinion 24-1 addressing lawyers' use of generative AI, confirming that client confidentiality obligations apply to data processed through AI tools. The California State Bar has proposed formal amendments to the Rules of Professional Conduct that would codify AI competence obligations directly into the disciplinary framework.

The direction is clear. The velocity is accelerating.

"The Tool Got It Wrong" Is Not a Defense

When a court imposes sanctions for AI-generated fabrications, it is not sanctioning the AI tool. The AI tool is not a licensed attorney. The AI tool has no duty of competence, no duty of candor, no duty to the tribunal. The tool cannot be disciplined, suspended, or disbarred.

The attorney can.

The attorney's signature on the filing is a certification — under Federal Rule of Civil Procedure 11, under state equivalents, under the court's inherent authority — that the attorney has conducted a reasonable inquiry and that the filing is well-grounded in fact and law. That certification does not say "to the best of my AI tool's knowledge." It is the attorney's personal representation.

Relying on an AI tool to do that verification — and accepting its output without independent confirmation — is not a reasonable inquiry. It is a delegation of professional judgment to a system that cannot exercise professional judgment.

The Competence Standard in Practice

ABA Formal Opinion 512 and the emerging state-level guidance converge on a practical framework. Here is what the competence obligation actually requires when you use AI in legal work:

Understand the tool before you use it. What does the tool do well? Where does it fail? What are its known limitations? A lawyer who cannot answer these questions has not satisfied the competence threshold for using the tool on a client matter.
Verify every output independently. AI-generated citations must be confirmed against primary sources — Westlaw, Lexis, the official reporter. Every single one. Not a sample. Not the ones that "look unusual." Every citation. Every time.
Apply your own legal judgment. AI can produce a draft. It cannot exercise professional judgment about whether the argument is sound, whether the authority is current, or whether the precedent applies to your specific facts. That judgment is yours. It cannot be delegated.
Document your verification process. Your documentation of the verification you performed is a defense. If you cannot show that you verified, you cannot demonstrate that you met the standard.
Evaluate your vendor's data handling. Before you use any AI tool on a client matter, review the vendor's terms of service, data residency provisions, model training policies, and confidentiality protections. If your client's confidential information enters a model training pipeline, that is a Rule 1.6 violation — regardless of whether you knew it was happening.

The Standard of Care Is Moving — and It's Moving Fast

In 2026, over thirty states have either issued formal AI ethics opinions, proposed rule amendments addressing AI, or published guidance on AI competence obligations. The ABA's Formal Opinion 512 has become the baseline. State bars are building on top of it.

The trajectory is toward formal codification. California's proposed rule amendments would make AI competence a disciplinary matter — not merely a best-practice recommendation, but an obligation enforceable through the state bar's disciplinary process. Other states are watching.

The attorney who is waiting for a final, definitive rule before building a verification practice is making a category error. The standard of care in professional responsibility has never required a formal rule — it requires what a reasonably competent attorney in similar circumstances would do. And in 2026, a reasonably competent attorney verifies AI output before relying on it.

That standard is already here. The formal rules are catching up to it.

What This Means for Your Practice, Right Now

You do not need to stop using AI. The attorneys being sanctioned are not being punished for using AI tools. They are being punished for using AI tools without verification — for treating AI output as finished work product rather than as a first draft requiring professional review.

The distinction matters. AI can make you dramatically more efficient — at research, at drafting, at synthesizing complex records. But efficiency without verification is not competence. It is speed on a road with no guardrails.

Build a verification checklist and use it every time.
Confirm every citation against primary sources before filing anything.
Review vendor terms before processing any client data through any AI tool.
Document your verification process.
Train everyone in your office on the same standard — the associate's filing carries your name if you supervise the work.

The attorney who doesn't verify AI is not just taking a risk. Under the current state of professional responsibility law, that attorney is not meeting the standard of care.

The attorney who doesn't read the cases before citing them is not a competent attorney. The attorney who doesn't verify AI output before filing it is the same attorney in a different era.

The era has arrived.

SavvyLex builds verification discipline into the workflow from the start. Vera, our governed legal AI assistant, is designed around citation hygiene, source verification, and audit-ready outputs — not as features, but as architecture. Every output is cited. Every source is verifiable. Every workflow leaves an audit trail.

See how it works: savvylex.com/vera