ABA Formal Opinion 512: What Generative AI Ethics Rules Actually Require of Attorneys

What Opinion 512 Is and What It Covers

ABA Formal Opinion 512 is the Standing Committee on Ethics and Professional Responsibility's first opinion focused specifically on generative AI — meaning large language model-based tools used to draft, research, summarize, or otherwise assist with legal work. It does not address AI broadly (e.g., document review platforms using earlier machine learning approaches), but focuses on the category of tools that generate novel text outputs from prompts.

The opinion does not create new rules. Instead, it maps existing Model Rules onto the specific risks that generative AI introduces. The committee identified five rules as directly implicated: Rule 1.1 (Competence), Rule 1.6 (Confidentiality), Rule 5.1 and 5.3 (Supervision of lawyers and nonlawyers), and Rule 3.3 (Candor toward the tribunal). Rule 1.5 (Fees) receives brief treatment in the context of billing for AI-assisted work.

Obligations by Rule: What the Opinion Actually Says

Rule 1.1 — Competence

The competence analysis in Opinion 512 has two layers. First, lawyers must understand the tool they are using well enough to deploy it competently — which the committee frames as understanding both the capabilities and the limitations of generative AI, including the documented tendency to produce plausible-sounding but factually incorrect output (commonly called hallucination). Second, lawyers must review AI-generated work product before relying on it, because the duty of competence cannot be delegated to a tool.

The opinion does not specify a minimum verification standard, but the committee's reasoning implies that blind submission of AI-generated content — without independent verification of citations, legal propositions, and factual claims — would fall below the competence threshold. The Mata v. Avianca sanctions decision (S.D.N.Y. 2023), though not cited in the opinion itself, sits as the obvious factual backdrop for this analysis.

Rule 1.6 — Confidentiality

This is where Opinion 512 gets most specific. Inputting client information into a generative AI tool — whether a general-purpose model like a public chatbot interface or a legal-specific platform — implicates Rule 1.6 unless the lawyer has taken reasonable steps to prevent unauthorized disclosure.

The committee identifies several questions lawyers should answer before using any AI tool with client data:

• Does the vendor's terms of service permit use of submitted data to train or improve the model?
• Where is data stored, and who has access to it?
• Does the vendor offer a zero-retention or enterprise data isolation option?
• Is the tool subject to a data processing agreement compatible with the attorney's confidentiality obligations?
• Has the lawyer obtained informed client consent where the data sharing arrangement warrants it?

The opinion does not categorically prohibit using cloud-based AI tools. It requires reasonable precautions — a standard that shifts based on the sensitivity of the client information involved. Highly sensitive matters (ongoing litigation, M&A, regulatory investigations) demand more scrutiny of the tool's data handling than routine administrative tasks.

Rules 5.1 and 5.3 — Supervision

Opinion 512 treats generative AI as analogous to a nonlawyer assistant for supervision purposes. Partners and supervising attorneys are responsible for establishing reasonable policies governing how AI tools are used within their organizations — and for ensuring that subordinate lawyers and nonlawyer staff follow those policies.

Practically, this means law firm AI policies are not just operational documents — they carry professional responsibility weight. A firm that deploys an AI tool without vetting its confidentiality practices, or that allows associates to submit AI-drafted briefs without review protocols, is potentially exposing supervising partners to Rule 5.1 liability, not just the associate who submitted the work.

Rule 3.3 — Candor Toward the Tribunal

The candor analysis is relatively brief but consequential. If an attorney submits AI-generated content containing fabricated citations or false statements of law, Rule 3.3 is implicated regardless of whether the attorney knew the content was false at the time of submission. The committee's position is that the competence duty under Rule 1.1 requires verification precisely to prevent Rule 3.3 violations — the two rules work together.

The opinion does not address whether attorneys must affirmatively disclose AI use to tribunals absent a court order or local rule requiring it. That question is left open, with the committee noting that disclosure obligations are governed by applicable court rules — an area where standing orders have proliferated significantly since 2023.

Rule 1.5 — Fees

The fee discussion is brief. The committee notes that billing a client for time saved by AI at the same rate as manual work — without disclosure — raises reasonableness concerns under Rule 1.5. It does not prescribe a specific billing model but flags the issue as requiring transparency with clients, particularly where engagement letters were negotiated before AI tools were deployed.

Obligations Summary

What the Opinion Does Not Address

Opinion 512 has meaningful gaps that practitioners should track through state bar guidance and court orders rather than assuming the ABA analysis is complete.

• Mandatory disclosure to clients of AI use: The opinion does not require it as a general matter, though it acknowledges some circumstances may warrant disclosure.
• Mandatory disclosure to courts: Left to applicable court rules. Attorneys should check standing orders in their courts — many federal district judges have issued AI disclosure requirements that go beyond what Opinion 512 requires.
• Unauthorized practice of law (UPL): Not addressed. Separate questions arise when AI tools are used in client-facing contexts without attorney oversight.
• Specific tool approvals or prohibitions: The opinion does not assess or endorse any named AI product.
• Agentic AI systems: Opinion 512 focuses on generative AI in a human-directed prompt-response model. Autonomous AI agents that take actions without per-step human instruction are not analyzed.

Practical Compliance Checkpoints

Opinion 512 does not provide a compliance checklist, but its analysis implies several concrete steps. The following maps the opinion's reasoning to operational actions:

1. Before adopting any AI tool: Review the vendor's terms of service and data processing agreement for training data use, retention periods, and access controls. Document this review.
2. For sensitive client matters: Confirm the tool offers enterprise data isolation or zero-retention mode before inputting client-identifying information. Where that is unavailable, consider whether the task requires client data at all.
3. For court filings: Independently verify every citation and legal proposition generated by the AI tool. Treat AI output as a draft that requires the same review you would apply to a first-year associate's work product.
4. At the firm level: Adopt a written AI use policy that addresses permissible tools, data handling requirements, verification protocols, and billing disclosure. Supervising partners should be able to point to this policy if a subordinate's AI use is questioned.
5. On billing: Review engagement letters to determine whether AI use materially changes the time or cost structure of the representation, and disclose accordingly.
6. For court practice: Check for standing orders on AI disclosure in each court where you practice. Opinion 512 does not displace court-specific requirements.

Relationship to State Bar Guidance

Opinion 512 arrived into a landscape where state bars had already begun issuing their own guidance. California's State Bar issued a framework document in late 2023 addressing similar competence and confidentiality concerns. The Florida Bar, New York City Bar, and Texas Bar have each issued formal or informal guidance that overlaps with — and in some cases goes further than — the ABA's analysis.

The relationship is additive, not hierarchical. An attorney in California is bound by California's Rules of Professional Conduct and the State Bar's guidance, not by the ABA Model Rules directly. Opinion 512 is useful for understanding the underlying reasoning, but it does not substitute for checking what the attorney's own jurisdiction requires.

Primary Source

The full text of ABA Formal Opinion 512 is available through the ABA Center for Professional Responsibility. Access requires ABA membership or purchase. The opinion identifier is Formal Opinion 512 (2024).