ABA Model Rules, with AI architected against them.
Privilege isn't a vendor promise. It's an architecture decision. We design legal AI deployments that satisfy ABA Model Rule 1.6, support Rule 1.1 competence, and align with the 2024-26 wave of state bar AI opinions.
Privilege is preserved by where the data flows.
The threshold question for any legal AI deployment is whether the AI processing pipeline constitutes disclosure to a third party. ABA Formal Opinion 512 and the state bar opinions that followed converge on the same answer: an AI deployment that keeps prompts and outputs within a single-tenant, contract- constrained, audited pipeline is not a waiver-creating disclosure.
We architect for the safe side of that line. Open-weights model inside the firm's perimeter; frontier model accessed only through dedicated tenancy with audited no-training and zero- retention; per-matter access controls; audit logs in the firm's own infrastructure.
The ABA Model Rules most relevant to legal AI deployments.
Confidentiality of information
A lawyer shall not reveal information relating to representation. AI processing of confidential information must not constitute prohibited disclosure — addressed through architecture (single-tenant, no-training, no-retention) and vendor contract.
Competence
A lawyer shall provide competent representation. Increasingly includes AI-tool competence: understanding capabilities, supervising output, verifying claims. Practice-area-specific competence applies on top.
Communication
Reasonable communication about the means by which the lawyer's objectives are to be accomplished. Some state bars now expect AI-use disclosure to clients on matters where AI plays a substantive role.
Fees
Reasonable fees; non-deceptive billing. Tool-driven efficiency can change the analysis on hourly billing for AI-assisted work. Several state opinions address this explicitly.
Responsibilities regarding non-lawyer assistance
Conduct compatible with professional obligations; reasonable efforts to ensure conformance. AI tools are non-lawyer assistance under this rule; the firm's supervision program must address them.
ABA Formal Opinion 512 (2024)
Comprehensive ABA guidance on generative AI in legal practice. Establishes the framework subsequent state bar opinions have built on. Reference document for our compliance maps.
The opinions converge on principles. They diverge on specifics.
Active or imminent AI-specific bar guidance in 2026 includes opinions from California, New York, Florida, Pennsylvania, Texas, Illinois, Virginia, and others. The principles broadly align with ABA Formal Opinion 512. The specifics — notice-to-client triggers, fee-bill disclosure expectations, generated-content marking — diverge.
For multi-jurisdiction firms, we produce a unified control set that satisfies the strictest applicable opinion across the firm's practice footprint. We don't design to the average; we design to the binding maximum.
For firms with overseas offices, GDPR (data transfer), the EU AI Act (where applicable), and Canadian PIPEDA + provincial bar rules layer on top. The same architecture supports them; the documentation and consent flows expand.
Frequently asked questions
Does using AI on privileged material waive privilege?
Not inherently. The waiver question turns on whether the AI processing constitutes disclosure to a third party. A properly-architected AI deployment — single-tenant, no-training contract, no-retention beyond purpose, audited — is not a waiver-creating disclosure under the ABA Formal Opinion 512 framework and subsequent state bar opinions. A consumer AI endpoint that may train on your inputs or that retains them in cross-customer logs is the opposite case.
What does Rule 1.1 require us to know about AI?
Rule 1.1 increasingly includes AI-tool competence. Practically: lawyers must understand the capabilities and limitations of AI tools they use, supervise output, verify factual claims, and protect against hallucinations and errors. The lawyer remains responsible for the work product regardless of which tool produced it. Practice-area-specific competence applies on top.
How do state bar opinions vary?
The 2024-26 wave of state bar opinions converged on the core principles (confidentiality protections, competence, supervision) but diverged on specifics: notice-to-client requirements, fee-bill disclosure of AI tool use, generated-content marking. We map state-specific requirements into the engagement architecture for firms practicing across multiple jurisdictions.
Can our paralegals use AI without firm sign-off on every tool?
The firm retains supervision responsibility, but practical governance can be tiered: a firm-approved tool list (post-vetting) gets routine use without per-matter sign-off; new tools require evaluation; per-matter sensitivity flags can elevate certain matters to attorney-only AI use. We help firms design the supervision policy alongside the technology.