In a post on X in September 2024, an engineer named Alex Bilzerian described a Zoom meeting with a venture capital firm that used Otter to record the call. After the meeting, he wrote, Otter automatically emailed him the transcript, "including hours of their private conversations afterward, where they discussed intimate, confidential details about their business."
These bots don't always introduce themselves, don't always know when the meeting is over, and can send the minutes to everyone on the list. The etiquette guides have caught up with the first problem: under "Be Upfront About Recordings," a U.S. Chamber of Commerce checklist by Beth Ziesenis starts with "Announce AI-assisted notetaking at the start of the meeting," then a disclaimer in the invite. The law can ask for more. The California Chamber of Commerce's September 18, 2026, alert reminds employers that the state is "a two-party consent state for audio recording," meaning everyone on the recording has to agree, so "employees must be notified and provide their consent before the tool is used to transcribe the meeting." Does a name in the participant list count as being told? Plaintiffs in a lawsuit against Otter give the example of a bot labeled "Anna's Notetaker (Otter.ai)."
A federal court in California, applying Washington law, has declined to assume that a visible bot amounts to notice. In that proposed class action, plaintiffs allege Otter's Notetaker joins meetings as a "silent participant." In an August 13, 2026, order that trimmed the case but kept its federal and California wiretapping claims alive, Judge Eumi K. Lee took up Otter's argument that a Washington plaintiff, Riley Dolan, had consented under Washington's privacy law because he could see the bot. "Even if the Otter Notetaker appeared in the meeting as another participant, the Court cannot infer that Dolan knew his communications were being recorded, transcribed, and retained by Otter," she wrote. That is early-stage reasoning, not a verdict, and the court dismissed his claim anyway because he hadn't alleged facts showing the conversations were private. Otter, for its part, denies the core notice allegations. Whatever the law finally decides, the etiquette is already settled: a name in the participant list is not an introduction.
The worse version is the bot whose owner is not even in the room. In September 2024 a former physician's Otter account joined an Ontario hospital's virtual rounds, a video call where doctors review their patients, more than a year after he had left, because the invite was still sitting in his personal calendar. That account comes from the hospital's breach report, as set out by Ontario's Information and Privacy Commissioner, which says the tool could "record the meeting without notice." The hospital found out only when the summary went out by email afterward, with a link to a transcript holding the health information of seven patients, sent to a recipient list of 65. It has since blocked AI scribe tools on site at its firewall and, in June 2025, revised its policy to tell staff to "review meeting participant lists for any inclusion of unapproved AI tools or automated agents, and remove them."
Otter's own guidance also recommends getting consent. Its help page says, "We always recommend getting consent from users before recording." Each user's own settings decide whether a warning email goes out, and making it company-wide takes the Enterprise plan; the email arrives five to ten minutes before the call.
One German regulator starts from a different place and in places goes further. Germany's Bavarian data protection authority, in its 2025 activity report, said that under the GDPR, Europe's data protection law, consent at work is often impractical because staff may not feel free to refuse. With contract usually unsuitable too in that setting, the questions it fielded were chiefly about legitimate interest, a legal basis under which the company weighs its own need against its staff's privacy in a way others can follow and check, and it set conditions. Switching transcription on wholesale or permanently is, as a rule, not warranted, and a company has to ask whether summary minutes or a person's notes would do. The report grants that those may work less well in long and complex meetings, in international teams, or where transcription helps people with disabilities take part. And anyone who objects mid-call must be dropped from the processing from that moment on.
So, the rules for hosts, in meeting order. Before: start where the strategist Amale Ghalbouni does in City AM, "Seek consent before introducing AI tools into meetings," and say where the notes will go and who gets them. Ask whether you need the bot at all, or whether a person's notes would do. Use a waiting room so you admit each participant yourself, as the U.S. Chamber recommends and Ontario's privacy office recommended to the hospital for meetings about patients. Can't make it? Ghalbouni again: "Ask to reschedule rather than send your bot in your place." During: if someone objects, switch it off for them. After: stop the bot before the people who stay behind talk among themselves, check who the summary will go to before it sends itself, and read the transcript before you quote it, since the California Chamber warns these tools "can also occasionally hallucinate details or misattribute statements." Then clear departed colleagues out of recurring invites.
And if you are the guest and an unannounced bot turns up, Ghalbouni says "don't hesitate to ask for it to be removed if it feels intrusive." One does not let a stranger sit at the table simply because he brought his own chair.
