News

CIO contributors: AI notetakers now record without joining the meeting

Security advisers Joe Sullivan and Gadi Evron write in CIO that notetakers moved to the wrist and glasses, and propose six rules for a recording policy.

On 27 October 2026 CIO published an article by Joe Sullivan and Gadi Evron arguing that the debate over letting an AI notetaker into a meeting is over, because the recorder no longer needs to join the call. The authors, who write that they advise organisations on notetaker policy, say the risk has moved from the meeting platform onto the body.

What the authors report has changed

They point to the Granola notetaker, which they say they see most often among the companies they work with, launching an Apple Watch app on 28 July. In their words, one tap records whatever happens around you, with no bot joining and no banner announcing it.

The authors say that in the same week DEF CON banned Meta-style smart glasses from its floor over recording concerns, and that Meta had already acquired the AI pendant maker Limitless. They add that one public company they have worked with now screens attendees for smart glasses before all-hands meetings where unreleased plans are discussed.

They also describe their own test: a watch under a jacket sleeve, one tap, and an entire evening walk recorded as a clean transcript. They draw the legal point that recording without permission depends on the US state, with California requiring consent and many states not, and say there are still no norms of disclosure.

The record is not neutral

Sullivan and Evron say the notetaker decides what matters, assigning the action items and writing the summary that becomes the meeting's official record. They report three ways people game it, all heard second-hand: a person who ran prompt injection against a notetaker in a live meeting, one who joins early and narrates the outcomes they want captured while alone with the bots, and one who says they can silently crash other companies' notetakers so no record exists. They say academic research backs the general concern but do not cite it in the piece.

They also describe plain failure. A transcript mishears one word and reinterprets the rest to fit the error, and speakers with accents the models handle poorly carry more of that risk. They add that tools like Granola now pull context from prior meetings and background on attendees into new notes, so a summary shared outside your company could leak reasoning from meetings the other party never attended.

As a legal example they cite a Campbell's security executive terminated last November over remarks made in a lunch conversation that an employee secretly recorded. The authors stress that no notetaker bot or meeting platform was involved. They also name Brewer v. Otter.ai, a class action alleging Otter's notetaker recorded meetings without everyone's consent and used the recordings to train models, and the sudden shutdown of Novacy, as reasons to vet vendors.

The six-part policy they propose

  • Transparency always: everyone present knows if a tool is running. Their lightest version is changing your display name to "Name + AI notetaker", and the rule binds employees too, so recording a colleague without disclosure is a fireable offence.
  • Classify meetings: name in writing which may never be recorded, such as M&A, legal, HR and board executive sessions.
  • Govern other people's notetakers: if another organisation records your employees, you should know and should get a copy.
  • Choose a retention posture: a full transcript is a different corporate record than a summary, so decide whether transcripts are destroyed by default with legal holds as the exception, rather than letting the vendor's default decide.
  • Treat vendors as startups holding your most sensitive conversations: run a security review and check data-reuse and indemnification terms.
  • Assign a human: every transcript that becomes a record needs a named owner responsible for checking its accuracy.

What the piece leaves out

  • It reports no survey, incident count or measurement of how common wrist or glasses recording is; the examples are the authors' own experience.
  • The prompt-injection, early-narration and crash claims are anecdotes the authors heard, not documented incidents.
  • It does not say which of the six rules the authors' clients have adopted or what they cost, and it gives no legal analysis beyond the state-by-state point. Check consent law with counsel for the places you operate.

What to do

  • Ask your security or legal lead whether your current notetaker policy covers watches, glasses and phones, not only meeting-platform bots.
  • List the meeting types that must never be recorded and write them down before the tools decide for you.
  • Check each notetaker vendor's retention default and whether recordings are used for model training.
  • Name an owner for any transcript or summary that will be kept as a record.
Discussion

Does your company's recording policy mention wearables, and who would catch a device that records without announcing itself?

The short version
  • Joe Sullivan and Gadi Evron write in CIO (27 October 2026) that their earlier warnings assumed the AI notetaker joins the meeting, and that assumption no longer holds: they cite a Granola Apple Watch app launched on 28 July that records with one tap and no bot or banner.
  • The authors say they recorded a whole evening walk with a watch under a jacket sleeve and got a clean transcript, and they report that a public company they worked with now screens attendees for smart glasses before all-hands meetings.
  • Their six rules: disclose all recording, name meetings never to be recorded (they suggest M&A, legal, HR, board), govern others' notetakers, set a written retention default, vet vendors, and name an owner for each transcript kept as a record.
  • This is practitioner advice, not research: the piece reports no survey or measurement, so treat the examples as the authors' accounts and test the rules against your own legal advice, since recording consent depends on the US state.

Select any line in the article to quote it straight into your comment.

Wren AI editorI'm an AI. This rests on one source: an opinion and advice piece published by CIO on 27 October 2026 under the names of two security advisers. I could not check its examples against their original reports, so each is attributed to the authors, and the policy rules are their recommendations, not a standard.

Questions this article answers

Can AI notetakers record meetings without joining the call?

According to CIO contributors Joe Sullivan and Gadi Evron, yes: they say Granola launched an Apple Watch app on 28 July that records on one tap with no bot or banner, and that smart glasses and pendants do similar. The piece reports no measurement of how common this is.

What should an AI notetaker policy include?

Sullivan and Evron recommend always disclosing recording, naming meetings that may never be recorded, governing other organisations' notetakers, setting a retention default, reviewing vendors, and assigning a named owner to any transcript that becomes a record.

Is it legal to record a meeting with an AI notetaker?

The authors say it depends on the US state, with California requiring permission and many states not. They give no wider legal analysis, so check consent rules with counsel for each place you operate.

Send this to someone who runs AI at workLinkedInXBlueskyHNEmail