Responsible AI Principles for Client-Facing Legal Work

Client-facing legal AI is categorically different from internal productivity AI. When AI is used to research a question faster or draft an email, the consequences of an error are manageable — the error can be caught in the drafting process. When AI outputs are transmitted to clients as legal advice or filed with courts as attorney representations, errors have professional responsibility consequences, client harm implications, and potentially disciplinary outcomes. The governance standards for client-facing AI must reflect this elevated risk.

Mandating attorney review for client work

Principle 1 — Human Review is Non-Negotiable: Every client-facing AI output must be reviewed by a qualified attorney before delivery. This is not optional under the ABA's supervision rules, and no volume of AI system reliability changes this requirement. The engineering design of client-facing AI systems should make review easy, visible, and traceable — not a step that can be skipped under time pressure.

Prioritizing precision over operational throughput

Principle 2 — Accuracy Over Speed: For client-facing work, a slower, more accurate output is always preferable to a faster, less accurate one. AI systems used for client-facing work should be calibrated for precision, not throughput. System design choices — prompt architecture, sampling strategy, model selection — should prioritize accuracy.

Designing systems to communicate limitations

Principle 3 — Uncertainty Must Be Surfaced: An AI system that generates confident-sounding text about a legal question it cannot reliably answer is a dangerous tool. Client-facing AI systems should be designed to surface uncertainty — to flag when a question is outside the model's reliable competence range, when the law is unclear or jurisdiction-specific, or when the available information is insufficient to support a definitive answer.

Establishing clear disclosure policies

Principle 4 — Disclosure Policies Must Be Clear: Firms must have a defined policy on when to disclose AI use to clients, consistent with state bar guidance in the relevant jurisdictions. The policy should not be left to individual attorney discretion — it should be part of the governance framework that every attorney using AI in client work understands and follows.

Maintaining traceability through audit trails

Principle 5 — Audit Trails Must Exist: For any client-facing AI output, the firm should be able to demonstrate what AI was used, what inputs were provided, what output was generated, and what attorney review was performed. This audit trail is both a quality control mechanism and a professional responsibility protection.