Key takeaways
- ABA Opinion 512 says informed client consent is required before entering information relating to a representation into a self-learning generative AI tool whose output could disclose it.
- Boilerplate consent in an engagement letter is not enough, according to the opinion.
- Read the tool's terms of use and privacy policy, or have someone qualified read them, before any client information goes in.
- Use of generative AI among legal professionals rose from 14% to 26% between Thomson Reuters' 2024 and 2025 reports, so firm rules are overdue.
- Many uses need no client data at all; separating those from the rest is the simplest protection.
Adoption is outrunning policy
Generative AI has moved from experiment to everyday tool in many firms. In Thomson Reuters Institute's 2025 report, 26% of legal professionals said they were already using generative AI, up from 14% in the 2024 report[1]. Training has not kept pace: a Thomson Reuters survey found that 64% of professionals had received no training on using generative AI for their industry work[2]. And clients are largely in the dark, with 71% of law firm clients saying they did not know whether their firms used generative AI tools[3].
Source: Thomson Reuters Institute, 2025[1]
Show the data
| Item | Value |
|---|---|
| 2024 report | 14% |
| 2025 report | 26% |
That combination, rising use, little training and limited client awareness, is where confidentiality problems happen. The fix is not to ban AI. It is to decide which uses are safe, which need safeguards and which need the client's consent, and to make those rules easy to follow.
What Opinion 512 requires
ABA Formal Opinion 512, issued in July 2024, applies the existing Model Rules to generative AI tools. On confidentiality, it says that before lawyers input information relating to a representation into a generative AI tool, they must evaluate the risk that the information will be disclosed to or accessed by others outside the firm, and also by others inside the firm who will not adequately protect it, for example because they do not know it came from a client[4].
Because many self-learning tools are designed so that their output could lead directly or indirectly to disclosure of client information, the opinion concludes that a client's informed consent is required before inputting information relating to the representation into such a tool[4]. For consent to be informed, the client must receive the lawyer's best judgment about why the tool is being used, specific information about the risks, including the kinds of information that will be disclosed and how others might use it against the client, and a clear explanation of the benefits[4]. Merely adding general, boilerplate provisions to engagement letters is not sufficient[4].
Not every use needs consent. Opinion 512 notes that if a lawyer uses a tool for idea generation without inputting information relating to the representation, client informed consent would not be necessary[4].
State guidance points the same way
State bars have reached similar conclusions. The State Bar of California's guidance says a lawyer must not input confidential client information into a generative AI tool that may present material risks to confidentiality or security without the client's informed consent[5]. Florida Bar Ethics Opinion 24-1 says lawyers may use generative AI but must protect client confidentiality, provide accurate and competent services, avoid improper billing and comply with advertising rules[6]. Check your own state's guidance, which may add specific requirements.
Settings and terms to check before use
Opinion 512 sets a baseline: all lawyers should read and understand the terms of use, privacy policy and related contractual terms of any generative AI tool they use, or consult a colleague or external expert who has, to learn who has access to the information entered[4]. It adds that lawyers may need to consult IT professionals or cybersecurity experts to understand those terms and how the tool uses information[4]. Use this checklist:
| Check | What to look for | Why it matters |
|---|---|---|
| Training on inputs | Contractual commitment that prompts and files are not used to train or improve models, or a setting that guarantees it | Training is the main route by which inputs could surface for other users |
| Data retention | How long prompts, files and outputs are stored, and whether the firm can delete them | Stored inputs are breach exposure and may need to be produced |
| Human review | Whether vendor staff can read inputs, for example for abuse monitoring | Another party with access to client information |
| Account type | Business or enterprise plan with admin controls, not personal accounts | Firm can enforce settings, MFA and offboarding |
| Sharing inside the tool | Whether chats, files or workspaces can be shared across users | Internal disclosure risk across matters and ethical screens |
| Subprocessors and location | Which other companies handle the data, and where | Required to judge who can access it |
| Security commitments | Encryption, access logging, breach notice terms | Same standard as any vendor holding client data |
Sort uses into three groups
- No client data: drafting a blog post outline, brainstorming deposition themes from public facts, summarizing a published opinion. Allowed on approved tools without consent.
- Client data in a vetted, non-training firm tool: summarizing documents or drafting from the file in a tool whose terms the firm has reviewed. Allowed under firm rules; consider whether the client should be told, and get consent where the tool's terms or the matter's sensitivity call for it.
- Client data in public or consumer tools: never, unless the firm has reviewed the tool and the client has given informed consent.
The same sorting applies to AI built into tools the firm already uses, such as email, document management and transcription, and to AI services that talk to prospective clients. If the firm uses an AI intake service such as Ava, put it through the same terms review as any other tool that handles confidential information.
A consent conversation that meets the standard
Where consent is needed, a short, specific conversation followed by a written confirmation works better than a clause buried in the engagement letter. Cover the points Opinion 512 lists[4]:
- Which tool, and why it helps in this matter, for example faster review of a large document set.
- What client information would be entered, and what would be kept out.
- The specific risks, including who could access the information and whether it could reach other users.
- What the firm has done to reduce those risks, such as a non-training enterprise plan.
- That the client may say no, and how the work would be done instead.
Writing the firm's AI use policy
A firm AI policy can fit on two pages. It should answer the questions staff actually have, in the order they have them:
- Which tools are approved, for what kinds of work, and who approved them.
- What may never be entered into any AI tool, such as privileged strategy in a consumer tool, other clients' information or anything under a protective order that limits disclosure.
- When a lawyer must get client consent, and where the consent record is kept.
- How outputs are checked before they are used, and who is responsible for the check.
- How AI use is recorded and billed.
- How to request a new tool, and who reviews it.
Review the policy whenever an approved vendor changes its terms or adds features, and at least once a year. Opinion 512 itself says its consent conclusion reflects the risks and capabilities of these tools as of its publication, and calls them a rapidly moving target whose features keep changing[4]. A policy written once will not stay accurate for long, so assign an owner and a review date.
Habits for every user
Use firm accounts only
Never paste client material into a personal AI account, even to save time.
Strip what is not needed
Remove names, account numbers and identifying details when the task does not need them.
Respect ethical screens
Do not reuse a shared workspace or chat history across matters where a screen applies.
Verify outputs
Check every citation and factual statement before relying on it; AI tools can produce confident errors.
Bill honestly
Opinion 512 says lawyers billing hourly must bill for actual time spent, not time saved by the tool[4].
Report surprises
If a tool shows information that seems to come from another user or matter, stop and report it.
Example: rolling out a document-summary tool
Example (fictional): a fifteen-lawyer commercial litigation firm wants to use an AI tool to summarize deposition transcripts. The firm's technology partner reviews the enterprise terms, confirms that inputs are not used for training and are deleted after a set period, and turns off sharing between workspaces so screened matters stay separate. The firm adds the tool to its approved list for client data, writes a one-page rule for associates, and for two clients in regulated industries holds a short consent conversation using the five points above, confirmed by email. Consumer AI accounts are blocked on firm devices.
Related guides
How long AI inputs and outputs should be kept is part of data retention and destruction policies. Train staff on these rules with security awareness training, see how AI is changing attacks in law firm cyberattack statistics and trends, and for client-facing AI on the phones, read using AI to answer after-hours intake calls ethically.
Frequently asked questions
Do lawyers need client consent to use AI tools?
Under ABA Opinion 512, informed consent is required before entering information relating to the representation into a self-learning tool whose output could disclose it. Uses that involve no client information, such as idea generation, do not need consent.
Is a clause in the engagement letter enough consent for AI use?
Opinion 512 says general, boilerplate provisions are not sufficient. Consent must be informed, with specific information about the tool, the information involved, the risks and the benefits.
Are enterprise AI tools safe for client data?
They can be appropriate after review. Check the terms on training, retention, human review, sharing and security, and keep a record of that review.
Can staff use free AI chat tools for client work?
Not with client information. Personal and free accounts usually lack the contract terms and admin controls a firm needs, and their terms may allow broader use of inputs.
References
Thomson Reuters Institute (2025). 2025 GenAI report: Executive summary for legal professionals (opens in a new tab) โ
Thomson Reuters (2025). Generative AI Adoption Nearly Doubles as Professional Services Reach Crossroads (opens in a new tab) โ
Thomson Reuters Institute (2025). 2025 GenAI report: Executive summary for legal professionals (opens in a new tab) โ
American Bar Association Standing Committee on Ethics and Professional Responsibility (2024). Formal Opinion 512: Generative Artificial Intelligence Tools (archived copy) (opens in a new tab) โ
The State Bar of California, Standing Committee on Professional Responsibility and Conduct (2026). Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law (opens in a new tab) โ
The Florida Bar (2024). Florida Bar Ethics Opinion 24-1 (opens in a new tab) โ