AI in law firms: confidentiality, Rule 1.6 and Opinion 512

Opinion 512 says the existing Model Rules apply to generative AI: competence (1.1), confidentiality (1.6), communication (1.4), meritorious claims and candor (3.1, 3.3, 8.4(c)), supervision (5.1, 5.3) and reasonable fees (1.5). The ABA Standing Committee on Ethics and Professional Responsibility issued it on July 29, 2024.

On competence, the opinion does not require lawyers to become AI experts. It requires a reasonable understanding of the capabilities and limitations of the specific tool, kept current as the tools change, which ties to Comment 8 to Rule 1.1 on the benefits and risks of relevant technology. Output must be reviewed to a degree that fits the tool and the task, and the lawyer remains fully responsible for the work.

The Model Rules are a template. Your state's adopted rules and its bar's ethics opinions are what bind you; Opinion 512 itself cites AI opinions from the Florida, Pennsylvania, West Virginia and California bars. Check yours before relying on this summary.

Under Opinion 512, informed consent is required before information relating to a representation is entered into a self-learning AI tool, because such a tool may later reveal it to other users inside or outside the firm.

Rule 1.6(a) bars revealing information relating to the representation without informed consent, implied authorization or an exception. Rule 1.6(c) requires reasonable efforts to prevent inadvertent or unauthorized disclosure or access. Comment 18 lists the factors: sensitivity of the information, likelihood of disclosure without added safeguards, the cost and difficulty of safeguards, and their effect on the lawyer's ability to represent the client.

For consent to be informed, the opinion says the client needs the lawyer's judgment on why the tool is used, the specific risks, what kinds of information will be disclosed and the benefits. It states that boilerplate in an engagement letter is not sufficient. Where no client information goes into the tool, for example pure idea generation, the opinion says consent is not needed.

For tools that do not learn from inputs, the opinion does not set a blanket rule. It asks for a fact-driven risk assessment that depends on the client, the matter, the task and the tool.

At minimum, read and understand the tool's terms of use, privacy policy and related contract terms, or consult a colleague or external expert who has. Opinion 512 calls this a baseline for every lawyer using a generative AI tool.

The opinion carries over checks from earlier cloud computing and outsourcing opinions. Rule 5.3 Comment 3 adds that the extent of the duty depends on the provider's reputation, the services, the terms protecting client information and the legal environment where the service is performed.

Not always: under Opinion 512, Rule 1.4 disclosure depends on the facts, but it is required if the client asks, the engagement terms demand it, Rule 1.6 consent is needed, AI affects the fee, or output influences a significant decision.

The opinion gives evaluating litigation outcomes or jury selection as examples of significant decisions. It also names the engagement agreement as a logical place to explain how the firm uses AI, even where no disclosure is required.

On fees, Rule 1.5 applies. A lawyer billing hourly charges for actual time, which includes the time to prompt the tool and to review the draft. A flat fee that stays the same although AI made the work much faster may be unreasonable. Tools that work like general office equipment are overhead; a per-use third-party charge for a specific client may be billed as an expense at actual cost. Time spent learning a tool the lawyer uses regularly may not be billed.

Managing lawyers must establish clear policies on permitted AI use, and supervising lawyers must make reasonable efforts to ensure that lawyers and nonlawyer staff follow them. Opinion 512 applies this to both the firm's own people and outside AI providers.

Training should cover the basics of the technology, the limits of the tools, the ethical issues and secure data handling. The opinion also suggests marking AI-generated material as such when it is stored in a client or firm file, so later users know it may be fallible. In a two-lawyer firm the policy and the training can be short, but both should exist.

The SRA's warning notice on misuse of AI, published 17 August 2026, says client information should only enter AI systems where appropriate contractual, technical and organisational safeguards are in place, citing paragraph 6.3 of both Codes of Conduct on confidentiality.

It asks firms to satisfy themselves that client data stays in a secure environment, is not accessed by unauthorised third parties, is not used to train AI models unless explicitly authorised and appropriate, and is not kept longer than necessary. It warns that both free and paid tools can store or reuse inputs depending on terms, settings and architecture, and that privilege may be lost. Supervisors remain accountable under paragraph 3.5 of the Code for Solicitors.

Data protection applies on top. Where an AI provider processes personal data for the firm, the ICO says UK GDPR requires a written contract that limits processing to the firm's documented instructions, requires security measures, controls sub-processors and provides for deletion or return at the end. The SRA's compliance tips add that it should always be clear to clients when they are interacting with AI.

A safe setup keeps client data inside the firm's own tenant, uses a provider contract that excludes training on inputs, puts a named lawyer's review before anything leaves the firm, and logs every step. These are design choices, not features you add later.

We build this way because it maps directly to the duties above. It does not replace your own assessment of your state's rules or the SRA's standards, and whether a specific tool is acceptable for a specific matter remains a professional judgment.