Lawyers use technology constantly without giving clients a tool-by-tool disclosure. Firms do not normally announce that a lawyer used Westlaw, spell-checking, document comparison software, e-discovery tools, or a spreadsheet.
Should generative AI be different?
The American Bar Association's Formal Opinion 512 does not create a universal rule requiring lawyers to tell clients every time generative AI is used. It says disclosure and informed consent may be required depending on how the tool is used, particularly where client information is being provided to systems that create confidentiality risks or where the client's instructions require communication. The ABA opinion also emphasizes competence, supervision, candor, and reasonable fees.
That gives firms room to use AI. It also means “we never disclose AI” is too simple a policy.
The tool itself is not always the material fact
Suppose a lawyer uses an approved enterprise AI system to summarize a long internal document, verifies the output against the original, and uses the summary only as a working aid.
The client may not need a special notification any more than the client needs to know which search interface the lawyer used.
Now change the facts.
The lawyer uploads confidential client material to a consumer AI service with uncertain retention terms. Or uses an AI system to perform a substantial part of an analysis the client expected a specialist to perform personally. Or passes a separate AI charge through to the client. Or agrees to a client guideline prohibiting specific AI tools.
In each case, AI use becomes more material.
The disclosure question should therefore follow the risk and the engagement, not the novelty of the software.
Confidentiality is the first test
Law firms should start with the information being processed.
MIRA's guide to attorney-client confidentiality emphasizes that the duty extends broadly across client-related information, including technology workflows. If a tool receives sensitive material, the firm needs to understand its security, retention, access, training, and contractual controls.
A client may reasonably care whether confidential information leaves the firm's controlled environment or becomes available to a third-party model provider.
Disclosure is particularly important when the lawyer cannot confidently conclude that the intended use is consistent with confidentiality obligations.
Billing is the second test
AI creates unusual billing questions because it can compress work.
If a client is billed hourly, the firm should not charge fictional time that the lawyer did not spend. If the firm charges separately for AI tools, the client should understand the basis for the charge. If the work is priced as a flat fee, the engagement should still comply with applicable reasonableness and communication requirements.
Accurate legal billing descriptions remain useful here. The narrative should tell the client what legal work was performed without exposing unnecessary confidential detail or using vague technology language to disguise the nature of the service.
The client is buying professional legal work. AI should not become a mechanism for making that work harder to understand.
Client instructions can settle the issue
Some clients will set explicit AI policies.
Corporate legal departments may prohibit certain consumer tools, require prior approval for processing confidential information, mandate specific security terms, or ask firms to disclose material AI use. Other clients may actively prefer AI-enabled workflows because they want lower costs and faster turnaround.
Those instructions should become part of matter governance.
A firm that can explain its AI architecture, review process, approved tools, confidentiality controls, and billing approach will be in a stronger position than one relying on an informal “everyone uses it now” assumption.
Routine AI will become less remarkable, accountability will not
Eventually, generative functions will be embedded so deeply into legal software that a binary “AI used: yes or no” disclosure may become meaningless.
The enduring questions are more practical.
Did the technology create a material confidentiality risk? Did it change the service the client reasonably believed it was buying? Did it affect fees? Did client instructions require disclosure? Was a lawyer still responsible for checking the work?
When the answer to one of those questions is yes, the client may have a legitimate right to know.
AI can become routine without becoming invisible where it matters.
Originally published on the MIRA blog.
Top comments (0)