Responsible AI for legal intake: control before scale.
AI intake sits at the front door of a professional service. It can reduce delay and administrative work, but it also receives sensitive facts before an attorney has reviewed the matter. Responsible deployment begins with a narrow scope, written controls, and accountable human oversight.
Define the permitted scope
Write down what the system may do and what it may not do. Permitted tasks may include identifying the caller, collecting approved facts, answering limited administrative questions, scheduling, routing, and summarizing. Prohibited tasks should include legal advice, legal conclusions, promises of representation, case-value estimates, and unsupervised conflict decisions.
Scope belongs in the workflow, staff training, vendor configuration, test scenarios, and quality review. A disclaimer cannot repair a system that is designed to cross the boundary.
Confidentiality and data minimization
Prospective clients may disclose highly sensitive information. The firm should understand what data the system collects, where it is transmitted, which subprocessors receive it, how long it is retained, whether it is used for model training, and how it can be exported or deleted.
Collect the minimum information needed for the intake decision. More data is not automatically better. Retention should be deliberate, access should be role-based, and vendor contracts should match the firm's obligations and risk tolerance.
Disclosure, consent, and recording
The firm should review applicable laws and professional rules for AI identification, call recording, automated communications, privacy notices, and consent. Requirements vary by jurisdiction, call direction, technology, and use. The approved greeting should be tested exactly as callers will hear it.
Inbound intake and outbound solicitation are not the same use case. Do not assume that a lawful inbound answering workflow authorizes automated outbound calls or messages.
Supervision and quality assurance
Assign a responsible attorney or operator. Review representative scenarios before launch and real interactions after launch. Verify summaries against recordings, confirm routing outcomes, investigate failures, and document material workflow changes.
ABA Formal Opinion 512 addresses lawyers' duties when using generative AI, including competence, confidentiality, communication, supervision, candor, and fees. Its specific application depends on the tool and jurisdiction, but the operating principle is clear: technology does not remove the lawyer's professional responsibilities.
Escalation and failure design
A responsible system is designed to stop, transfer, or flag when confidence is low or facts fall outside the approved path. Include human-requested transfers, urgent safety issues, potential conflicts, language uncertainty, caller distress, integration failures, and repeated misunderstanding.
Every transfer needs a fallback. Every incomplete interaction needs a visible status. Every material uncertainty needs to survive into the record.
This guide is a practical operating framework, not legal advice. Firms should evaluate applicable professional rules, privacy requirements, recording laws, and communications laws in each jurisdiction where they operate.
Sources and further reading
Primary and industry sources used to support this page. External guidance should be reviewed in context and for your jurisdiction.
- ABA Formal Opinion 512ABA guidance on lawyers' professional obligations when using generative AI.
- ABA, Law's New First ImpressionDiscussion of AI-supported client intake and human oversight.
- ABA formal ethics opinions indexOfficial index for current ABA formal opinions.