FRCP 11 and AI-Assisted Filings: The Certification You're Required to Make Every Time
Every attorney who files a brief containing AI-generated research has already made a Federal Rule of Civil Procedure 11 certification about that research. Most attorneys understand the rule exists. Fewer have mapped exactly what it requires in an AI-assisted workflow — and what ABA Formal Opinion 512 adds on top of it.
The professional responsibility framework governing AI use in legal practice is not waiting to be written. It is already written. Federal Rule of Civil Procedure 11(b)(2) has governed legal contentions in federal court since 1983. ABA Formal Opinion 512 (July 29, 2024) extended the existing ethics framework — competence, supervision, candor, confidentiality — to AI tools explicitly. The question is not whether these rules apply to AI-assisted work. They do. The question is whether attorneys have mapped their existing workflows against what the rules actually require.
Most have not, because the sanctions that have made headlines present the issue as a technology problem: AI hallucinated, attorney was sanctioned, lesson learned about AI. The framing obscures a prior step. Before the AI hallucinated, the attorney made a professional certification. Understanding what that certification requires — in a workflow where AI is generating the underlying research — clarifies what a responsible AI-assisted practice actually looks like.
The Certification: FRCP 11(b)(2)
Federal Rule of Civil Procedure 11(b) provides that by presenting a signed pleading, written motion, or other paper to the court, an attorney certifies that to the best of their knowledge, information, and belief, formed after an inquiry reasonable under the circumstances:
"the claims, defenses, and other legal contentions are warranted by existing law or by a nonfrivolous argument for extending, modifying, or reversing existing law or for establishing new law"
Source: Fed. R. Civ. P. 11(b)(2); law.cornell.edu/rules/frcp/rule_11
This certification is made on the attorney's signature. It is not made by the AI tool. The certifying attorney assumes professional responsibility for the accuracy of the legal contentions in the filing — including contentions that were generated, researched, or drafted with AI assistance.
The "reasonable inquiry" standard means the certification is not merely a subjective belief — it is an objective test applied to what a reasonable attorney would have done before signing. Courts evaluating sanctions under Rule 11 ask whether the attorney conducted the inquiry that the circumstances required. An attorney who uses an AI-generated case citation without independently verifying that the case exists has not met this standard, regardless of how plausible the citation looked.
What the Inquiry Requires for AI-Generated Research
Courts have not issued a uniform standard for what a "reasonable inquiry" means specifically in the AI-assisted context. But the sanctions decisions issued to date — beginning with Mata v. Avianca, Inc. and followed by similar rulings in multiple federal districts through 2025 and 2026 — converge on a consistent minimum. Before filing any brief that relies on AI-generated research, the following steps are not optional.
1. Confirm the case exists. Pull the full opinion from a primary source — a court's official database, Westlaw, LexisNexis, or a free primary source like CourtListener or Google Scholar. If you cannot locate the case independently, do not cite it.
2. Verify the holding you attributed. Read the passage of the opinion your brief relies on. Confirm that the proposition you are citing is actually what the court held — not a paraphrase of what the AI predicted the court might have said.
3. Check subsequent history. Confirm the case has not been overruled, reversed on the cited point, or limited by later decisions in controlling jurisdictions. An accurate cite to an overruled case is still a false statement of current law for FRCP 11 purposes.
4. Document your steps. Maintain a file note that records what you verified, when, and how. This creates a contemporaneous record that the inquiry was conducted, which becomes critical evidence if sanctions are later sought.
None of these steps are new. They are the steps a careful attorney applies to all research. The AI context adds a specific failure mode — confident-sounding fabricated citations — that makes independent verification more consequential, not less.
ABA Formal Opinion 512: The Competence Framework
ABA Formal Opinion 512 (July 29, 2024), issued by the ABA Standing Committee on Ethics and Professional Responsibility, addresses the application of the Model Rules of Professional Conduct to attorneys who use generative AI tools in legal practice. The opinion does not prohibit AI use. It applies the existing framework — Rules 1.1, 1.4, 1.5, 1.6, 3.3, 5.1, and 5.3 — to the AI context explicitly, and identifies six professional duties that the combination of those rules generates.
Competence (Rule 1.1): Attorneys must understand the AI tools they use well enough to evaluate their output critically — including understanding common failure modes like citation hallucination. Using a tool you do not understand well enough to verify is not competent use.
Supervision (Rules 5.1, 5.3): Supervising attorneys are responsible for AI-generated work used by those under their supervision, the same way they are responsible for work product generated by associates or paralegals. AI tools are not self-supervising.
Communication (Rule 1.4): Whether to disclose AI use to a client is a context-sensitive judgment, but attorneys should consider whether the client would want to know. The opinion notes that some engagement agreements now address AI use explicitly.
Candor (Rule 3.3): An attorney who submits a brief containing AI-generated false statements of law without verifying them may have violated Rule 3.3(a)(1) — the prohibition on knowingly making false statements to a tribunal. An attorney who does not verify AI-generated citations may satisfy the knowing element if they did not check what they were on notice to verify, but the rule requires knowledge, not strict liability.
Confidentiality (Rule 1.6): Transmitting client information to a third-party AI service raises confidentiality obligations. Attorneys must review the service agreement and understand how client data is used by the provider before using commercial AI tools with client information.
Billing (Rule 1.5): If AI tools reduce research time, billing the client for equivalent time as if the research were done manually may be a fee reasonableness problem. The opinion does not resolve this question but identifies it as one attorneys should address.
Source: ABA Formal Opinion 512 (July 29, 2024), ABA Standing Committee on Ethics and Professional Responsibility. Source: americanbar.org.
The Asymmetry AI Introduces
The practical consequence of ABA FO 512 in an AI-assisted workflow is counterintuitive: using AI to generate research can increase the verification burden rather than reduce it. A careful attorney who performs their own research develops a natural sense of whether a source is reliable — the source is familiar, the holding aligns with what the attorney already knows about the area, the citation pattern is consistent. An attorney reviewing AI-generated output does not have that baseline. The output may look exactly like reliable research even when it is fabricated, because language models are trained to produce text that sounds like reliable research.
This means the inquiry required before an AI-assisted filing is not the same kind of inquiry a careful attorney would apply to their own research. It is a more adversarial inquiry: assume the output contains errors, and design the verification process to catch them. Every citation verified, every holding confirmed against the primary source, every subsequent history checked.
The attorneys who have been sanctioned were not all careless. Some were caught by citations that looked entirely plausible. The FRCP 11 and ABA FO 512 framework does not make exceptions for plausibility. The certification is absolute, and it belongs to the attorney.
What This Means for Legal AI Design
The framework points to a specific standard for any AI tool intended for professional legal use. An AI tool that surfaces a legal exception with a citable primary source, states the confidence level of its analysis, and flags what it could not verify is a tool that supports rather than undermines an attorney's FRCP 11 and ABA FO 512 compliance. A tool that generates confident-looking citations without stating its basis is a tool that creates certification problems, not one that solves research problems.
Legal Exception is built to meet the first standard: surface the exceptions and exemptions that are actually there, cite them to primary sources, and stop rather than speculate where the source is unavailable. That is what a responsible AI-assisted legal research tool should do — not because it is commercially convenient, but because the rules require no less.
The content on this site 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 situation.
References & Sources
- Federal Rule of Civil Procedure 11(b)(2) — certification requirement for legal contentions in federal court filings; "reasonable inquiry" standard. Source: law.cornell.edu/rules/frcp/rule_11.
- ABA Standing Committee on Ethics and Professional Responsibility, Formal Opinion 512 (July 29, 2024) ("Generative Artificial Intelligence Tools") — applies ABA Model Rules 1.1, 1.4, 1.5, 1.6, 3.3, 5.1, and 5.3 to attorney use of AI tools; establishes competence, supervision, candor, and confidentiality duties. Source: americanbar.org.
- ABA Model Rules of Professional Conduct, Rule 1.1 (Competence — including duty to keep abreast of relevant technology and its benefits and risks); Rule 3.3(a)(1) (Candor toward the tribunal — prohibition on making false statements of law to a court); Rules 5.1 and 5.3 (Responsibilities regarding nonlawyers and support staff, extended to AI tools under FO 512). Source: americanbar.org.
- Mata v. Avianca, Inc., No. 22-cv-1461 (S.D.N.Y. June 22, 2023) — foundational AI-hallucination sanctions order; six fabricated citations in a brief generated using ChatGPT; court imposed sanctions on both attorneys and their firm. See also subsequent federal district court orders (2025–2026) requiring attorneys to certify AI citations were independently verified before submission.
- Federal Rule of Civil Procedure 11(c) — sanctions mechanism; court may impose appropriate sanctions on an attorney who violates Rule 11(b); standard is objective reasonableness, not subjective bad faith. Source: law.cornell.edu/rules/frcp/rule_11.