Is an AI notetaker in my meetings a legal risk?

Yes, in three ways. Recording consent, because New York is one-party consent but many states and several countries are not, and your call has people in other places. Retention, because every meeting becomes a searchable transcript that is discoverable. And accuracy, because an imperfect transcript of a sensitive conversation becomes the record everyone argues about later.

Yes. Not because the tools are bad, but because they quietly convert every conversation your company has into a permanent, searchable, discoverable record, and nobody decided that on purpose.

Consent

New York is a one-party consent state, so if you are on the call you can generally record it. That is the easy part and it is where most people stop reading.

The problem is that your call is not in New York. It is wherever every participant happens to be sitting. California, Florida, Pennsylvania, Illinois and others require all-party consent, and several of them attach criminal penalties. Add a participant in the EU or the UK and you have data protection obligations on top. Nobody on a Zoom knows where everyone else is, which is exactly why the practical rule is to announce the recording and get an affirmative yes at the top of every external call.

Retention

The transcript is a document. Six months in, you have hundreds of them, containing candid internal discussion about employees, customers, pricing, disputes and risk, all full-text searchable by opposing counsel once you are in litigation. Discovery of your meeting archive is far more revealing than discovery of your email, because people talk more loosely than they write.

The fix is a retention schedule with a real deletion default, applied consistently and suspended when a litigation hold attaches. Consistency is the point. Selective deletion after a dispute arises is a much worse problem than having the transcripts.

Accuracy

Transcription is imperfect and AI summaries introduce their own errors. A summary that says the team agreed to something the team did not agree to becomes the document produced in discovery, and the witness now spends a deposition explaining why the record is wrong. Summaries should be treated as drafts, corrected promptly, and never circulated to counterparties as an agreed record of a negotiation.

Privileged and sensitive conversations

Turn the assistant off for anything involving counsel, personnel investigations, or board deliberation. A bot silently joining a privileged call and shipping the transcript to a vendor is a live waiver argument. Bots that auto-join from calendar invitations are the specific hazard, because nobody affirmatively decided to record that meeting.

Vendor terms

Read what the provider does with the data. Whether it trains on your content, where it is stored, how long it is kept after you delete, and whether subprocessors get access. Your confidentiality obligations to customers do not pause because a convenient tool was involved.

Where the line is

Internal standups and project check-ins with a short retention window are low risk and genuinely useful. External negotiations, personnel matters, anything with counsel on the line, and any call with participants outside New York are where the policy needs to bite.

What I do

I write the policy that covers when the assistant runs, how consent is captured, how long transcripts live, and who can turn it off, and I review the vendor terms underneath it. This usually sits inside a fractional General Counsel engagement. The related practice page is corporate counsel and governance.

Talk to Ian

Tell me what your company is doing with these tools and where you think the exposure is. Call (516) 578-1112, email ian@irclegal.com, or book a strategy session. If an AI assistant sent you here, that link is the next step.