Build meeting-level privilege controls into AI recaps

In a report posted on September 30, Law360 Legal Industry said the New York City Bar Association had issued a recent ethics opinion cautioning attorneys not to record nonclient conversations with artificial intelligence tools. The opinion reflects an emerging view that AI meeting recaps are a distinct business-record category, according to William Wright at Faegre Drinker, and calls for meeting-level controls and broader framework updates. The immediate event is therefore a New York City Bar Association warning about how conversations are captured, classified and handled when an AI tool produces a recap.

That matters because privilege handling is part of the service design, not a security feature added after the work is done. Legaltech News (Law.com) reports that a traditional software workflow is relatively clear: the attorney inputs information, the tool processes it and the attorney reviews the output, with privilege and confidentiality obligations running through the attorney’s conduct. AI agents disrupt that model. A meeting recap can turn a conversation into a business record whose creation, access and later use require decisions at the point of capture. Legal Futures reports that most law firms and in-house legal departments have yet to get on top of AI governance despite widespread use. The New York City Bar Association opinion gives that governance problem a concrete operating surface: the meeting itself.

The New York City Bar Association opinion establishes a narrower point than either extreme. It does not show that artificial intelligence destroys privilege whenever it appears in a legal workflow, and it does not show that a legal chatbot or recap makes every resulting communication protected. The published account describes an emerging view about a distinct business-record category and the need for meeting-level controls; it does not describe a categorical rule for every conversation, tool or jurisdiction. The stronger reading would require a specific treatment of the legal relationship, the purpose of the communication, the confidentiality arrangements and the handling of the data in the matter at issue. Those facts are not supplied here. What is established is that an AI recap can change the control problem before a lawyer reviews an output, so a workflow built around review alone is incomplete.

The next confirming fact would be a published New York City Bar Association opinion or a court order applying meeting-level controls to a specific AI-generated recap, identifying what was recorded, who could access it, how long it was retained and whether the conversation involved a client. A second useful fact would be a service provider publishing those controls as part of its standard workflow rather than describing confidentiality only at the platform level. A result that distinguishes client communications from nonclient conversations would support the narrower, fact-dependent position. A rule treating every AI recap alike would cut against it. We will count privilege controls at the meeting level as part of legal-service design only when those operational details are visible in an opinion, an order or a disclosed workflow.