ABA Rule 3.3 and AI Citations: What "Knowingly" Means When Your Research Comes From a Language Model

Most of the legal discussion around AI hallucinations and attorney sanctions has focused on Federal Rule of Civil Procedure 11. That focus is understandable — Rule 11 has been the vehicle for most of the sanctions orders that have made headlines, and its "reasonable inquiry" standard is an objective test that does not require proving intent. But running parallel to Rule 11 is a separate obligation that receives less attention: ABA Model Rule 3.3(a)(1), the candor-to-tribunal provision, which requires a different mental state — actual knowledge — and carries different consequences.

Understanding Rule 3.3's "knowing" standard in the AI context matters because the two frameworks diverge in important ways. An attorney can violate FRCP Rule 11 without subjective bad faith — the test is objective reasonableness, and a sincerely held but unreasonable belief about a citation's accuracy is still a Rule 11 violation. Rule 3.3(a)(1), by contrast, requires that the attorney knew the statement was false. What "knowing" means when the attorney never verified whether the AI-generated citation was real is a question that remains under-analyzed, and the answer is more complex than it first appears.

Rule 3.3(a)(1): The Text and Its Requirements

ABA Model Rule 3.3(a)(1) provides that a lawyer shall not knowingly make a false statement of fact or law to a tribunal, or fail to correct a false statement of material fact or law previously made to the tribunal by the lawyer. The rule contains two distinct obligations: a prohibition on affirmatively making false statements, and an ongoing duty to correct false statements previously made. Both obligations apply to legal authority, not just factual assertions.

The ABA Model Rules define "knowingly" in Rule 1.0(f): "Knowingly," when used in reference to a lawyer's conduct, denotes actual knowledge of the fact in question. A person's knowledge may be inferred from circumstances. That final sentence — "knowledge may be inferred from circumstances" — is where the AI context raises the most significant interpretive question.

The "Knowingly" Standard: When Actual Knowledge May Be Inferred From Circumstances

Rule 3.3(a)(1) requires actual knowledge, not mere negligence. An attorney who genuinely did not know that an AI-generated citation was fabricated has not "knowingly" made a false statement in the traditional sense. That is why Rule 3.3 sanctions are much rarer than Rule 11 sanctions in the AI-hallucination cases that have emerged — it is harder to establish that an attorney knew the citation was false when the attorney believed, however unreasonably, that the citation was real.

But the definition of "knowingly" includes the critical qualification that knowledge may be inferred from circumstances. It is this inference of actual knowledge from circumstances — not a lower negligence standard — that makes the Rule 3.3 analysis more difficult for attorneys who use AI tools without adequate verification practices. The relevant circumstances may include:

These circumstances bear on whether actual knowledge can be inferred — they do not substitute negligence for knowledge. An attorney who genuinely did not know a citation was false, and where no reasonable inference of knowledge arises from the circumstances, has not violated Rule 3.3(a)(1), even if that attorney was negligent in failing to verify. Negligence is the domain of FRCP Rule 11, not Rule 3.3.

Willful blindness — deliberately avoiding knowledge of a fact — is treated as equivalent to actual knowledge in most legal contexts. An attorney who has been put on notice of AI hallucination risk, understands that AI tools fabricate citations, and deliberately skips verification to save time is not meaningfully different, in terms of culpability, from an attorney who knows a citation is false. Whether bar disciplinary authorities and courts will apply the willful-blindness doctrine in AI hallucination cases is an open question, but the doctrinal framework supports it.

The Duty to Correct: Rule 3.3(a)(1)'s Second Obligation

Rule 3.3(a)(1) also imposes a duty to correct false statements of law previously made to the tribunal. This obligation is ongoing — it does not expire when a brief is filed. If an attorney discovers, after filing, that an AI-generated citation was fabricated, Rule 3.3(a)(1) requires the attorney to correct that false statement. The correction must be made regardless of whether the false statement affected the outcome, and regardless of whether opposing counsel noticed the error.

The correction obligation is one of the most practically significant aspects of Rule 3.3 in the AI context. An attorney whose AI-generated brief contained fabricated citations and who discovers the fabrication post-filing faces a clear professional obligation: notify the tribunal that the cited authority does not exist. The obligation to correct arises at the moment the attorney gains actual knowledge of the falsity — which may be when opposing counsel challenges the citation, when the attorney independently checks the source, or when the court notes the error. At that point, silence is not an option under Rule 3.3(a)(1).

How ABA Formal Opinion 512 Frames Rule 3.3

ABA Formal Opinion 512 (July 29, 2024) identifies Rule 3.3 as one of the rules directly implicated by AI use in legal practice. The opinion's treatment of Rule 3.3 is measured: it notes that an attorney who submits AI-generated authority without verification "may have" violated Rule 3.3(a)(1), depending on whether the knowing standard is met. The opinion does not definitively resolve when actual knowledge may properly be inferred from circumstances, but it does establish that the competence duty under Rule 1.1 — which requires understanding the AI tool's limitations — is a predicate for assessing Rule 3.3 compliance.

The connection matters: an attorney who satisfies the Rule 1.1 competence duty by understanding that AI tools hallucinate citations, and who then files those citations without verification, is not operating from ignorance. The attorney who understood the risk and skipped verification is in a different position from the attorney who had no idea that AI tools fabricate case names. Rule 3.3's "knowing" standard tracks that distinction, and ABA FO 512's competence framework means that the ignorance defense weakens over time as AI hallucination becomes more widely documented.

The Practical Difference From FRCP Rule 11

The most important practical difference between FRCP Rule 11 and ABA Rule 3.3 is the enforcement mechanism and consequences. Rule 11 sanctions are awarded in the civil litigation where the violation occurred and are limited to what suffices to deter — typically monetary sanctions, though the court may impose other appropriate remedies. Rule 3.3 violations are professional responsibility matters that may be reported to the state bar, resulting in disciplinary proceedings that can include suspension or disbarment.

An attorney sanctioned under Rule 11 for filing AI-generated fabricated citations has a civil record of the sanctions order. That same attorney, if a state bar disciplinary authority concludes that the conduct also violated Rule 3.3, faces the possibility of professional consequences that extend far beyond a single case. The distinction matters most in cases where the hallucination conduct was egregious — multiple fabricated citations, repeated after a prior warning, or filed in a context where the attorney's lack of verification was particularly reckless.

This content is legal information, not legal advice. It does not create an attorney-client relationship and cannot substitute for consultation with a licensed attorney about your specific circumstances.