**ABA Model Rule 5

ABA Model Rule 5.3 requires lawyers to supervise nonlawyer assistants — and in 2026, that includes AI tools. ABA Formal Opinion 512 (2024) explicitly mapped Rule 5.3's supervision requirements to AI, establishing that attorneys must oversee AI output with the same diligence they'd apply to work from a junior associate or paralegal.

The practical implication is straightforward: if you use an AI tool to draft a motion, conduct research, or review documents, you're responsible for the output as if you'd delegated the task to a first-year associate who's never been to court. The AI doesn't understand what it's producing. You do. That gap is what supervision is designed to close.

Rule 5.3: What It Actually Says and How It Applies to AI

Model Rule 5.3 requires that partners and supervising lawyers ensure that nonlawyer assistants' conduct is compatible with the professional obligations of the lawyer. Originally written for paralegals, legal assistants, and investigators, the rule has always been broader than its original context.

ABA Formal Opinion 512 extended the rule to AI tools by analogy. The opinion doesn't call AI a "nonlawyer assistant" in the literal sense — AI isn't a person. But the supervision framework applies because the risk is identical: work product generated by something other than a licensed attorney enters the legal process, and a lawyer must ensure it meets professional standards.

Under this framework, the supervising attorney must:

Understand the tool's capabilities and limitations — what it does well, where it fails, and how to distinguish reliable output from hallucination. This is the competence requirement (Rule 1.1) meeting the supervision requirement (Rule 5.3).

Review all output before it's used — every citation, every legal proposition, every factual claim. Not spot-checking. Comprehensive review, the same way you'd review a first-year's research memo before relying on it.

Maintain responsibility for the final work product — the attorney's name is on the filing, and the attorney bears the consequences. "The AI wrote it" is not a defense, just as "my paralegal wrote it" has never been a defense.

Supervising AI Like a Junior Associate

The analogy to supervising a junior associate is imperfect but useful. Here's how the parallel works — and where it breaks down.

What's the same: You assign a task with clear parameters. You review the output before using it. You verify the research independently. You make the judgment calls about strategy and analysis. You take responsibility for the final product.

What's different — and harder with AI: A junior associate learns from feedback. AI doesn't improve within a matter based on your corrections. A junior associate flags uncertainty — "I'm not sure about this holding." AI states fabricated citations with the same confidence as verified ones. A junior associate understands the purpose of the task. AI processes tokens.

This means supervision of AI must be more rigorous than supervision of associates, not less. The associate who says "I found a case that's directly on point" has probably found something relevant. The AI that says the same thing may have invented the citation entirely.

The practical standard: treat every piece of AI output as a first draft from someone who's smart but has no judgment. It might be excellent. It might be fabricated. You won't know until you verify. Every time.

What Supervision Means in Practice: Workflow Requirements

Supervision isn't a concept — it's a workflow. Here's what it looks like operationally:

For AI-assisted research: The attorney defines the research question. AI generates an initial research memo with citations. The attorney verifies every citation in Westlaw or Lexis — confirming the case exists, the holding matches, and the authority is current. The attorney independently assesses whether the analysis is correct. Only then does the research enter the work product.

For AI-assisted drafting: The attorney outlines the arguments and identifies the key authorities. AI generates a first draft. The attorney reviews every sentence for accuracy, completeness, and appropriateness. The attorney revises the draft to reflect their professional judgment, not just the AI's suggested framing. The attorney confirms that the final product is one they'd be willing to sign.

For AI-assisted document review: The attorney defines the review criteria and relevance standards. AI flags potentially relevant documents. The attorney reviews a statistically significant sample to verify AI accuracy. The attorney makes final determinations on privilege, relevance, and production — AI flags are inputs, not decisions.

For all AI use: Document what tool was used, what task it performed, and what verification steps were taken. This creates an audit trail that demonstrates supervision was exercised — not assumed.

Documentation Requirements: Proving You Supervised

Supervision only counts if you can prove it happened. When a filing is challenged, when a client complains, or when a bar disciplinary committee investigates, the question will be: "What supervision did you exercise over the AI output?"

What to document:

The AI tool used and the specific task assigned. The prompt or instructions given to the AI. The AI's output (save it — don't just take what you need and discard the rest). The verification steps taken, including which citations were checked, which sources were consulted, and what changes were made. The attorney who conducted the review and when.

How to document:

Some firms build this into their document management systems — every AI interaction gets logged in the matter file. Others use simpler approaches: a verification checklist appended to every AI-assisted filing, a brief memo in the matter file noting AI use and verification steps.

The standard isn't perfection — it's reasonable supervision. If you verified every citation, reviewed the analysis, and applied professional judgment to the final product, you've met the standard even if an error slips through. If you didn't verify, didn't review, and filed AI output as-is, no amount of documentation saves you.

Practical tip: Create a standard verification form for your firm. Tool used, task assigned, output reviewed by, citations verified by, date completed. Takes 60 seconds to fill out per task and creates a defensible record.

Partners' Liability Under Rules 5.1 and 5.3 Combined

Rules 5.1 and 5.3 create a two-layer liability structure that managing partners need to understand.

Rule 5.1 makes partners responsible for ensuring that all lawyers in the firm comply with ethics rules. If an associate uses AI irresponsibly and the firm has no AI policy, no training, and no oversight, the managing partner is exposed under Rule 5.1 — not for using AI, but for failing to establish systems that ensure compliance.

Rule 5.3 makes the supervising attorney responsible for the nonlawyer assistant's conduct. When AI is the "assistant," the attorney who relies on AI output is the one who bears responsibility for verification and quality.

Combined, these rules mean: The managing partner must establish AI governance policies and training (Rule 5.1). The supervising attorney must verify AI output in each matter (Rule 5.3). The associate who uses AI must exercise competence in how they use it (Rule 1.1).

The cascade of liability is real. When *Mata v. Avianca* happened, it wasn't just the filing attorney who faced consequences — the supervising partner was sanctioned too. Courts look at the entire chain of supervision, and "I didn't know he was using AI" doesn't absolve the partner who should have had systems in place to manage the risk.

The investment required to comply is modest: a written AI policy, annual training, enterprise AI tools, and a verification workflow. The cost of non-compliance — sanctions, malpractice claims, bar discipline, and reputational damage — is exponentially higher.

The Bottom Line: Rule 5.3 means every attorney who uses AI must supervise its output like a junior associate's work — verify every citation, review every analysis, document the process, and accept full responsibility for the final product, because 'the AI wrote it' has never been and will never be a defense.