ABA Formal Opinion 512: The lawyer's new duties in the AI era

On July 29, 2024, the American Bar Association's Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 512: Generative Artificial Intelligence Tools. It was the first comprehensive ABA opinion specifically addressing generative AI in legal practice.

It is the most consequential ethics guidance of the AI era.

Formal Opinion 512 does not prohibit lawyers from using generative AI. It does something more important: it identifies six distinct duties every lawyer using these tools now owes — duties grounded in the existing Model Rules of Professional Conduct, but with new specific applications to AI workflows.

Every state bar has now incorporated some version of these duties into its own guidance. If you practice law and use AI, you are operating under this framework whether you have read it or not.

Here is what each duty requires, and what it means in practice.

Duty 1 — Competence (Model Rule 1.1)

The competence rule has always included a duty to maintain understanding of the technology used in your practice. Formal Opinion 512 makes explicit what this means in the AI context.

A lawyer using generative AI must understand:

• That the tool is capable of producing false information that looks authoritative
• The general nature of how the tool produces output (without needing to understand the underlying mathematics)
• The limitations of the tool in legal contexts specifically
• When the tool's output requires independent verification (which, for citations to authority, is always)

You don't have to be a computer scientist. You do have to understand enough to make informed judgments about what the tool can and cannot reliably do. A lawyer who treats a chatbot as a search engine — believing its outputs are authoritative simply because they sound authoritative — is not competent to use it.

In practice: every lawyer using AI should be able to articulate, in two or three sentences, how their tool works and where it can fail. If you can't, you shouldn't be using it on client matters yet.

Duty 2 — Confidentiality (Model Rule 1.6)

Most lawyers using consumer AI tools are violating the confidentiality rule without realizing it.

When you paste a deposition transcript into a consumer AI tool, that text is transmitted to the provider's servers. Depending on the service tier and configuration, it may be retained, used for model training, or accessible to the provider's personnel. The same is true across most consumer AI products.

Formal Opinion 512 is clear: confidential client information cannot be transmitted to a generative AI tool unless:

• The lawyer has obtained informed consent from the client (Rule 1.6(a))
• OR the tool's terms of service and technical architecture provide adequate confidentiality protections that satisfy the lawyer's duties

Most consumer AI products do not satisfy the second prong. Most lawyers using them have not obtained explicit informed consent for AI processing. The result is a widespread, often unrecognized violation of Rule 1.6.

A lawyer using AI for legal work should be using a tool with:

• Contractual confidentiality protections (zero-retention terms, no training on customer data)
• Encryption in transit and at rest
• Ethical wall enforcement at the infrastructure layer
• Audit logs showing exactly what data was processed and when

In practice: if you can't point to the specific contractual and technical protections your AI tool provides, assume you're violating Rule 1.6.

Duty 3 — Communication (Model Rule 1.4)

A lawyer must keep clients reasonably informed about the means used to handle their matter. Formal Opinion 512 extends this to the use of generative AI in two specific situations:

1. When AI use is "material" to the representation — for example, where AI is being used to draft pleadings, conduct research, or analyze documents
2. When the client has expressed a preference about AI use or has a legitimate interest in knowing

In practice, this means:

• Engagement letters should address AI use (your firm's general practice, with options for client preferences)
• Significant AI-generated work product should be transparent — clients should know it was AI-assisted
• Billing should accurately reflect AI use (more on this under Duty 6)

The opinion stops short of requiring affirmative disclosure in every case. But it warns that lawyers who use AI without any client communication and without any documented confidentiality protection are exposed.

Duty 4 — Candor toward the tribunal (Model Rule 3.3)

This is the duty that Mata v. Avianca violated, and it is the duty that produces sanctions.

Rule 3.3 prohibits a lawyer from knowingly making a false statement of fact or law to a court, and requires the lawyer to take reasonable remedial measures if false evidence has been offered. Formal Opinion 512 makes explicit that:

• Citing a fabricated case generated by AI violates Rule 3.3 — even if the lawyer believed the citation was real
• The duty applies whether the lawyer or the AI produced the false content
• "I didn't know it was wrong" is not a defense; the duty includes a duty to verify

This is why every state bar's AI guidance now includes some version of the same instruction: every citation produced by AI must be independently verified before it appears in a court filing. Not after. Before.

In practice: a firm using AI for drafting needs a verification step that is structural, not discretionary. If verification depends on the individual lawyer remembering to do it, it will eventually be skipped and someone will get sanctioned.

Duty 5 — Supervision (Model Rules 5.1 and 5.3)

Rule 5.1 covers supervision of subordinate lawyers. Rule 5.3 covers supervision of non-lawyer assistants. Formal Opinion 512 makes clear that generative AI tools fall within the scope of Rule 5.3.

This has several consequences:

• Partners and managing lawyers must establish policies governing AI use within the firm
• Those policies must be reasonably designed to ensure AI use complies with all applicable rules
• Failure to establish or enforce AI policies can itself be a Rule 5.3 violation
• Supervising lawyers can be held responsible for AI-generated errors by associates if the firm's AI policies were inadequate

In practice, every firm using AI needs:

• A written AI policy
• Training for everyone who uses AI tools
• A documented verification process for AI outputs
• A mechanism for surfacing AI errors to supervisors

Firms that haven't done this work are exposed. The duty doesn't depend on whether the firm has had an AI incident yet. The duty is to have the supervision infrastructure in place before something goes wrong.

Duty 6 — Reasonable fees (Model Rule 1.5)

Rule 1.5 requires that lawyer fees be reasonable. Formal Opinion 512 addresses AI's impact on billing in two directions.

First, lawyers cannot bill for AI-generated time as if it were lawyer time. If JILL drafts a motion in a fraction of the time it would have taken an associate, the client cannot be billed for the associate hours that were never spent. The opinion is clear: the time billed must reflect the time actually expended.

Second, lawyers should not bill clients for AI subscription costs as separate line items unless agreed to in the engagement. The opinion treats AI tools the same way it treats research database subscriptions — generally a cost of doing business, not a separately billable expense.

In practice, AI is going to compress the time required for many legal tasks. The fee question is how the firm captures the value of that compression. Lawyers who try to bill hourly for AI-compressed time are exposed under Rule 1.5. Lawyers who shift to flat-fee or value-based pricing for AI-assisted work are aligned with the rule.

Why these six duties matter as a system

Each of these six duties stands alone. But the deeper point of Formal Opinion 512 is that they operate together.

A lawyer who is competent (Duty 1) to use AI but does not maintain confidentiality (Duty 2) has not satisfied her professional responsibilities. A lawyer who maintains confidentiality but doesn't supervise her associates' AI use (Duty 5) has not satisfied them either. A lawyer who supervises well but bills inflated AI time (Duty 6) is still violating the rules.

The framework is cumulative. To use AI ethically, the lawyer must satisfy all six duties — and in practice, satisfying any one of them depends on having infrastructure that supports the others.

This is why ad-hoc AI use, with each lawyer making her own choices about which tool to use and how to verify, is structurally untenable. The duties are systemic. They require systemic answers.

What a firm-level answer looks like

A firm that satisfies all six duties at the system level has:

• A single AI platform with documented competence training for every user (Duty 1)
• Contractual and technical confidentiality protections built into the platform infrastructure (Duty 2)
• Engagement letter language addressing AI use and client preferences (Duty 3)
• Structural verification of every AI output against authoritative sources, before the output reaches the lawyer (Duty 4)
• Written AI policies, training records, and audit logs accessible to supervisors (Duty 5)
• Billing practices aligned with actual time expended and the AI tool's role in the work (Duty 6)

This is not a compliance checklist. It is the architecture of a defensible AI practice.

Most firms today are not there. They are using consumer AI tools, in unstructured ways, without documented policies, without confidentiality protections, and without verification infrastructure. They are exposed under at least three of the six duties, often all six.

The path to compliance with Formal Opinion 512 is not "use AI more carefully." The path is to use AI tools that were built to satisfy these duties from the ground up — and to retire the consumer tools that were not.

That is the architectural argument Discover Docket exists to make. DDEAS — our Ethical and Accountability Standards framework — was designed to satisfy each of these six duties by default, at the infrastructure layer, in a way that does not depend on the individual lawyer remembering to do the right thing.

Formal Opinion 512 told the profession what was now required. Discover Docket is what compliance with it actually looks like in practice.

---

Chris Waters is a 25-year California civil litigator and the founder of Discover Docket. Related essays in this series cover Mata v. Avianca (the canonical AI sanctions case) and the architectural design of the DDEAS framework.