ABA Opinion 512 One Year Later

ABA Formal Opinion 512 gave U.S. lawyers a clear frame for generative AI: competence, confidentiality, supervision, communication, and candor still apply — the technology does not create new exceptions. A year into broader firm adoption, the divide is no longer between firms that allow AI and firms that ban it. It is between firms that operationalized the opinion and firms that issued a one-page memo and hoped for the best.

Firms that moved early share a few patterns. They maintain an approved tool list with security review status. They separate tools that can touch client data from tools that cannot. They require human review of any AI-assisted work product before it leaves the building. And they train in the workflow — not as a CLE checkbox — so associates know what may be pasted where.

Confidentiality remains the sharp edge. The most common failures are not dramatic hallucinations — they are quiet ones: a summer associate pasting a draft into a consumer chatbot, a knowledge-management bot retaining prompts without a retention policy, or an integration sending metadata to a vendor with weak subprocessors.

Second: supervision. Reviewing AI output requires domain competence and skepticism. Junior lawyers rubber-stamping generated citations — and partners rubber-stamping juniors — recreate the classic malpractice chain with a new accelerant.

Third: inconsistent client positions. One client is told AI is never used; another is told it is used everywhere. Inconsistency becomes an exhibit if a dispute arises.

A defensible program under Opinion 512 looks like this in discovery: a written policy with dates and owners; evidence of training completion; approved tool inventory with security diligence; matter-level usage rules; sample review checklists; and a documented incident path for suspected disclosure failures. If you cannot produce those artifacts, your program is a slogan.

Disclosure and consent standards vary by state bar guidance and by use case. Treat disclosure as a risk-based decision: higher sensitivity when AI substantially shapes advice or filings without meaningful human contribution; lower when AI is used as a drafting accelerant under close supervision. Document the rationale either way — silence without analysis is the worst position.