The ABA Said Attorneys Must Understand AI. Here's What That Actually Requires.

In August 2024, the American Bar Association issued Formal Opinion 512 — the most significant formal ethics guidance the ABA has issued on technology in the modern era of AI.

The opinion didn’t say attorneys must use AI. It said attorneys must understand it — and that using AI without understanding its limitations, data handling practices, and appropriate verification requirements may constitute a competence violation under Model Rule 1.1.

That’s a meaningful shift. The professional obligation is no longer just about whether you use AI correctly. It’s about whether you understand it well enough to use it at all.

What ABA Formal Opinion 512 Actually Says

The full opinion runs 22 pages. The practical obligations it establishes fall into four categories:

Competence (Model Rule 1.1): Attorneys must understand the technology they deploy in client matters, including AI tools. “Understanding” is defined functionally — not requiring technical expertise in how large language models work, but requiring knowledge of what a specific tool does, what it doesn’t do reliably, what its hallucination profile is, and what verification is required before relying on its output.

Supervision (Model Rules 5.1 and 5.3): Partners and supervising attorneys are responsible for ensuring that subordinates — including associates and paralegals using AI tools — use them competently. Delegating to AI doesn’t eliminate supervisory responsibility. If an associate uses an AI tool to draft a brief and the supervising partner doesn’t understand enough about that tool to review the output appropriately, the supervision obligation may not be met.

Confidentiality (Model Rule 1.6): Client data submitted to AI tools must be handled in compliance with confidentiality obligations. This creates specific requirements around tool selection — tools that store client data, use it for model training, or share it with third parties create confidentiality exposure that the attorney is responsible for managing.

Candor (Model Rule 3.3): AI-generated content submitted to tribunals must be accurate. The Mata v. Avianca case — where fabricated AI citations were submitted to a federal court — resulted in sanctions precisely because the attorneys violated candor obligations. Verification of AI output is not optional for anything submitted to a court or opposing counsel.

The State Bar Landscape in 2026

The ABA issues guidance; state bars regulate. The state bar responses to AI have been uneven but are consolidating toward consistent themes.

State

AI Ethics Action

Key Requirement

New York

Mandatory CLE

2 credits/year in AI competency starting 2025

California

Formal guidance issued

Competence, confidentiality, disclosure requirements

Florida

Formal guidance

Verification of AI output before submission

Texas

Advisory issued

Client consent recommended for AI use on matters

Illinois

Working group formed

Rules update in progress

Colorado

Formal guidance

Similar to ABA Opinion 512 framework

The trajectory is clearly toward AI competence as a baseline professional obligation, not a specialty skill. New York’s mandatory CLE requirement is the leading indicator — other states will follow as AI adoption increases and bar associations see more AI-related misconduct.

The Disclosure Question

One area where bar guidance is still developing: whether attorneys must disclose AI use to clients.

The current consensus is nuanced:

No disclosure required for AI tools used in the same way a legal research database or word processor is used — as a tool that assists the attorney’s work without changing the attorney-client relationship.

Disclosure recommended or required in three specific situations:

• When AI is being used to reduce time in a matter billed by the hour, and the reduction affects what the client pays
• When client confidential data is being processed by a third-party AI system
• When the client has specifically asked about AI use in their matter

Some states are moving toward stronger disclosure requirements. Texas’s guidance recommends client consent for AI use on matters, which goes further than most other states. The safest approach for firms operating across multiple jurisdictions is an AI use disclosure that covers the variations.

What “AI Competence” Actually Requires in Practice

The ABA’s competence standard doesn’t require attorneys to understand transformer architecture. It requires them to understand what a specific tool does in legal practice contexts.

For any AI tool used in a matter, the attorney should be able to answer:

1. What data does this tool process, and where is it stored?
2. Does this tool’s provider use submitted data for model training?
3. What is this tool’s known failure rate for the specific task I’m using it for?
4. What verification protocol applies to this tool’s output before I rely on it?
5. Does this tool’s training data cover the jurisdiction and practice area I’m applying it to?

These aren’t technical questions. They’re questions about professional responsibility. The answers determine whether using the tool in a specific context is competent representation.

The Supervision Gap

The supervision obligation creates a specific practical problem for law firm management.

Under Model Rules 5.1 and 5.3, partners and supervising attorneys are responsible for ensuring that supervised lawyers and non-lawyer staff comply with professional conduct rules. If a paralegal is using an AI tool to prepare immigration petitions or draft discovery responses, the supervising attorney must be capable of supervising that use.

Supervising AI use requires:

• Knowing which tools are in use
• Understanding what those tools do and where they can fail
• Having a review process that catches AI errors before they reach clients or courts
• Ensuring non-lawyer staff using AI are operating within defined parameters

This is why centralized AI policy — knowing what tools are approved, who can use them, and what review applies — is not just good operational practice. It’s a professional conduct requirement for supervising attorneys.

Building a Compliant AI Policy: The Minimum Viable Framework

Only 32.9% of law firms have established any AI use policy (AllRize 2025). For the 67% that haven’t, the professional conduct risk is sitting unmanaged.

A compliant AI policy doesn’t need to be a treatise. It needs to cover:

Tool approval: Which specific tools are approved for which use cases. Prevents attorneys and staff from running client data through consumer-facing AI tools that don’t meet confidentiality standards.

Data classification: What categories of client information can be processed through which tools. Highly sensitive matters (privilege logs, trade secret materials, sealed documents) should require more restrictive data handling.

Verification requirements: What review is required before AI output is used in client deliverables or court filings. The standard can vary by use case — a research summary requires different verification than a brief or a contract redline.

Disclosure protocol: How and when AI use is disclosed to clients and in court filings. This includes firm-specific decisions on the disclosure threshold that apply consistently across all attorneys.

Training requirement: Who must complete training before using approved tools, and what that training covers. Training should address both the tool’s capabilities and the firm’s specific workflows and verification standards.

The Candor Standard in Practice: What Courts Are Doing

Following Mata v. Avianca, courts have become more proactive about AI-related candor issues. Several federal districts have issued standing orders requiring disclosure of AI use in filings. The landscape as of early 2026:

Several federal district courts now require disclosure of whether AI was used to draft or research court submissions, and whether that output was reviewed for accuracy by a licensed attorney. State courts are following in jurisdictions with high-volume litigation.

The practical implication: any attorney using AI in litigation should assume that disclosure requirements apply or will apply, and that the verification standard for AI-generated content submitted to courts is review by a licensed attorney who can take responsibility for its accuracy. This is not a new standard — it’s the existing standard for any work product submitted to a court. AI doesn’t reduce it.

The Malpractice Dimension

Professional ethics obligations and malpractice liability are related but distinct. ABA guidance establishes what competent representation requires. Malpractice turns on whether a breach of that standard caused a client harm.

As AI adoption increases, the malpractice exposure landscape shifts in two directions:

Liability for AI-related errors: Attorneys who use AI tools and fail to verify output may face malpractice claims when that output is incorrect and the client suffers harm. The Mata scenario — submitting fabricated citations — is the most obvious example, but the risk extends to any AI output that reaches a client without appropriate verification.

Liability for AI non-adoption: As AI tools become standard, firms that fail to use available tools efficiently may face questions about whether the time and cost they incurred was reasonable. If a task that competent attorneys routinely complete in one hour using AI tools takes five hours without them, the question of whether the additional billing was appropriate may arise.

This second risk is emerging rather than established. But its trajectory is clear as AI competence becomes a baseline expectation.

Frequently Asked Questions About AI Ethics Compliance

Does ABA Formal Opinion 512 require me to use AI?
No. It establishes competence and confidentiality obligations for attorneys who choose to use AI. It does not require adoption. It does say that if you use AI, you must understand it well enough to use it competently.

What happens if an attorney uses an unauthorized AI tool with client data?
Potential consequences include confidentiality violations under Model Rule 1.6, disciplinary proceedings, and malpractice exposure if the data handling results in client harm. The appropriate protection is a firm policy that defines approved tools and prohibits use of consumer-facing tools for client matters.

Is client consent required for AI use?
Currently, the ABA and most state bars do not require consent for AI tools used in standard legal workflows. Texas’s guidance recommends consent — which may indicate where national guidance is heading. The safest approach is disclosure in the client engagement letter that AI tools may be used in the representation.

How should AI-generated content be verified before filing?
The verification standard is the same as for any other work product: a licensed attorney must review it, take responsibility for its accuracy, and be capable of certifying its accuracy to the tribunal. The AI generates a draft. The attorney certifies the final product. There is no shortcut on certification.

Where can I find state-specific AI ethics guidance?
Paxton.ai maintains a comprehensive 50-state survey of AI ethics guidance. The ABA’s Center for Professional Responsibility maintains formal opinions. State bar websites publish formal ethics opinions. The picture varies significantly by state — a multi-state practice should track each jurisdiction’s requirements separately.

The Bottom Line

ABA Formal Opinion 512 didn’t create new professional obligations from scratch. It clarified that existing obligations — competence, supervision, confidentiality, candor — apply to AI tools the same way they apply to every other tool in legal practice.

The difference is that most attorneys can supervise a junior associate’s research. Fewer can currently supervise an AI tool’s output with the same confidence. That’s the gap that training, policy, and workflow design closes.

The firms that treat AI ethics compliance as a professional obligation — not just a risk management exercise — are the ones building practices that can grow confidently as AI becomes standard across legal work.

NexLink helps law firms build AI governance frameworks that satisfy professional conduct requirements — including policy development, attorney training, verification workflow design, and compliance documentation across the ABA and state bar standards that apply to your practice.

Sources:

• ABA Formal Opinion 512 — Generative Artificial Intelligence Tools
• 2025 State Bar Guidance on Legal AI — Paxton
• AllRize 2025 Legal Technology and AI Adoption Report
• New York CLE AI Competency Requirement — NY State Bar
• The Legal Industry Report 2025 — American Bar Association
• Mata v. Avianca — AI Sanctions Case Analysis
• 2025 Clio Legal Trends Report