AI Adoption in Law Firms: What the 2024 Survey Data Says About Attorney Competence Obligations

The gap between how many attorneys are using AI tools and how many have formal verification practices in place widened noticeably through 2024. Survey data from the American Bar Association's annual Legal Technology Survey Report and Thomson Reuters' State of the Legal Market report both documented adoption rates climbing past 35–40% among responding attorneys — while the share reporting any structured review protocol for AI-generated output remained substantially lower.

That gap matters because it sits directly inside the competence obligation. ABA Model Rule 1.1, Comment 8 states that competent representation requires keeping abreast of changes in the law and its practice, including the benefits and risks associated with relevant technology. A majority of state bars have adopted or substantially tracked that comment. What it means operationally — how much an attorney needs to understand about a tool before deploying it on client work — has become the central professional responsibility question of the current adoption cycle.

What the 2024 Adoption Data Actually Shows

Several surveys published in 2024 and early 2025 attempted to quantify attorney AI use. They differ in methodology and sample composition, which makes direct comparison difficult — but the directional findings are consistent enough to draw working conclusions.

The variance across these figures is partly definitional. "Using AI" in one survey may mean a weekly interaction with a legal research assistant; in another it captures any use of a general-purpose chatbot for work-adjacent tasks. The Thomson Reuters figures, which focus on purpose-built legal AI tools rather than general LLMs, likely provide a more operationally meaningful baseline for competence analysis.

Where Competence Obligations Attach

The professional responsibility question is not whether attorneys may use AI — virtually every state bar opinion published through 2024 has confirmed they may, subject to conditions. The question is what conditions apply and at what point in the workflow.

At least 20 state bars had published formal ethics opinions or guidance on AI use by the end of 2024, with California, Florida, New York, and Texas among the most detailed. While the specific language varies, a common structure has emerged across these opinions:

• Competence (Rule 1.1): Attorneys must understand the tool's capabilities and limitations sufficiently to supervise its output. This does not require technical expertise in how the model works, but does require knowing what kinds of errors it produces and at what rate.
• Confidentiality (Rule 1.6): Client data submitted to AI tools implicates confidentiality obligations. Attorneys must assess the vendor's data retention and training practices before using client information in prompts.
• Supervision (Rules 5.1 / 5.3): Partners and supervising attorneys bear responsibility for subordinates' AI use. Firm-level policies on AI review and verification are increasingly treated as a supervision obligation, not merely a best practice.
• Candor (Rule 3.3): Submitting AI-generated citations without independent verification has produced the most visible sanctions to date. Several courts have now issued standing orders requiring attorneys to certify that AI-generated citations have been verified.

The Verification Gap: Where Adoption Has Outrun Practice

The documented incidents of AI-related sanctions through 2024 share a common pattern: an attorney used a generative AI tool to assist with research or drafting, the tool produced plausible-looking but fabricated citations or case summaries, and the attorney submitted the output without independent verification. The Mata v. Avianca matter (S.D.N.Y. 2023) established this pattern in public record; subsequent incidents have followed the same structure.

What the 2024 survey data adds is scale context. If roughly 35–40% of US attorneys are using AI tools in practice, and if a meaningful fraction of that group is not applying systematic output verification, the documented incidents represent a small but growing subset of a much larger unverified-use population.

Firm Size and the Verification Divide

Large firms have moved faster on formal AI governance than small ones — not because large-firm attorneys are more cautious, but because they have the infrastructure to build it. AmLaw 100 firms began publishing internal AI use policies in 2023 and 2024, typically covering tool approval processes, data classification requirements before use, and mandatory human review of AI-generated work product before client delivery.

Solo practitioners and small firms are in a different position. They have adopted AI at meaningful rates — the ABA survey shows solo practitioners among the more active early adopters, likely because efficiency gains are more immediate when there is no associate to delegate to — but formal verification protocols are far less common. A solo practitioner using a general-purpose LLM for initial research drafts and then verifying manually is operating within a reasonable competence framework. One who is not verifying is not.

Practice Area Variation

Adoption is not uniform across practice areas, and the competence risk profile varies accordingly.

What Bar Opinions Have Actually Required

The state bar opinions published through 2024 have been more specific than many attorneys realize. Several have moved beyond general competence language to address particular workflows.

An attorney using a generative AI tool must understand the technology sufficiently to evaluate its output, recognize its limitations, and determine when human judgment must supplement or replace the AI's work product. Reliance on AI output without such evaluation does not satisfy the competence obligation under Rule 1.1.

Florida Bar Ethics Opinion 24-1 (2024) — paraphrased; consult primary source for exact language

California's State Bar issued a formal report in November 2023 that, while not a binding ethics opinion, laid out a detailed framework that has influenced subsequent guidance. It explicitly addressed the confidentiality implications of submitting client data to cloud-hosted AI tools without zero-data-retention agreements, and noted that the competence obligation requires attorneys to understand whether a tool uses submitted data for model training.

New York's City Bar issued guidance in 2024 that addressed supervision obligations specifically — noting that a supervising partner who directs a junior associate to use AI tools without providing guidance on verification procedures may bear responsibility for resulting errors under Rule 5.1.

The Confidentiality Dimension: Data Retention and Tool Selection

A thread running through most 2024 bar guidance is the data retention question. General-purpose LLMs accessed through consumer interfaces typically retain conversation data and may use it for model improvement. Purpose-built legal AI tools marketed to law firms generally offer zero-data-retention configurations, enterprise agreements with explicit training exclusions, or on-premises deployment options.

The practical implication: an attorney who uses a consumer ChatGPT interface to draft a memo containing client-identifying details is in a different position from one using a firm-provisioned tool with a documented zero-retention policy. The former requires either client consent or careful scrubbing of identifying information before use. The latter still requires verification of output, but the confidentiality analysis is cleaner.

Survey data from 2024 suggests that a substantial fraction of attorney AI use is happening through personal or general-purpose accounts rather than firm-provisioned tools — particularly at smaller firms and among solo practitioners. This creates a confidentiality exposure that most attorneys using these tools have not formally analyzed.

Court Orders and the Certification Trend

Federal district courts began issuing AI-specific standing orders in 2023, and by mid-2024 dozens of courts had adopted some form of AI disclosure or certification requirement. The specific requirements vary — some require disclosure of AI use in filings, others require certification that AI-generated content has been reviewed for accuracy, and a few require disclosure of the specific tools used.

These orders operate independently of bar ethics obligations. An attorney who violates a court's AI certification order faces potential sanctions under Fed. R. Civ. P. 11 or the court's inherent authority, regardless of whether the bar separately pursues a disciplinary matter. The Mata sanctions — which included a $5,000 fine and mandatory CLE requirements — were imposed under the court's inherent authority and Rule 11, not through a bar disciplinary proceeding.

What Adequate Competence Looks Like in Practice

Bar opinions and court orders establish the obligation; they do not specify the exact verification steps. Based on documented incidents and the published guidance available through early 2025, a working framework for competent AI use in legal practice includes the following:

1. Tool-level due diligence before first use: Understand the tool's data retention policy, whether client data is used for training, and what the vendor's accuracy benchmarks are for your specific use case. This is a one-time investment per tool, not per matter.
2. Citation verification as a non-negotiable step: Every case citation generated by an AI tool must be independently verified in a primary legal database before filing or client delivery. No exception based on the tool's stated accuracy rate.
3. Output review calibrated to stakes: A first-pass research memo for internal use warrants different review intensity than a brief filed in federal court. The verification protocol should scale with the consequence of error.
4. Confidentiality screening before prompt construction: Determine whether the tool's data handling is compatible with submitting client-identifying information. If not, anonymize or use a tool with appropriate contractual protections.
5. Supervision documentation at the firm level: Firms with associates using AI tools should have a written policy specifying what review is required before AI-assisted work product reaches a supervising attorney or client. The absence of such a policy is itself a supervision risk.

The Trajectory Into 2025 and 2026

The adoption curve shows no sign of flattening. Purpose-built legal AI tools have continued to expand their feature sets — moving from isolated research assistance into integrated drafting, contract lifecycle management, and litigation analytics. As tools become more capable and more embedded in standard workflows, the competence question shifts from "do you know how to use this" to "do you have firm-level governance that keeps the human-in-the-loop obligation operational at scale."

Several state bars have signaled that updated or more specific guidance is forthcoming. The ABA's Center for Professional Responsibility has been tracking the issue and has indicated that formal model rule commentary updates or a formal opinion on AI use are under consideration. When those are issued, they will likely clarify — rather than relax — the verification and supervision requirements that have emerged from state-level guidance.

For practitioners, the practical question is not whether to use AI but whether the current workflow would survive scrutiny under Rule 1.1 and the applicable bar guidance in their jurisdiction. The documented incidents suggest that a meaningful number of current practices would not.