Who’s Listening?

There's a new participant joining our meetings, and increasingly, it's not a person.

AI meeting assistants have quietly become part of everyday working life. They join calls, record conversations, produce transcripts, summarise discussions and identify actions. They promise to make meetings more productive, remove the burden of taking notes and ensure nothing important gets missed. On the surface, it's difficult to argue with the benefits.

You might have carefully assessed the AI tools your organisation uses, completed a privacy impact assessment, reviewed the provider's security arrangements and established rules around recording and retention. Then you join a meeting with a customer, supplier or consultant, and their AI assistant appears alongside them, or worse, in their place.

Suddenly, your organisation's carefully considered approach to privacy may have very little bearing on what happens to the conversation.

The assistant might announce that it's recording or transcribing or simply appear as another participant. Either way, most people carry on with the meeting. Nobody wants to be the person who interrupts to question the presence of a note-taking tool, particularly when everyone else seems comfortable with it.

But what exactly are we agreeing to when we allow someone else's AI assistant to listen to our conversations?

We probably don't know who is processing the information or for what purposes, where the transcript will be stored, who will have access to it or how long it will be retained. We may not know whether the information can be used to improve the provider's services, whether recordings are retained alongside summaries, or whether the transcript might be shared with people who weren't even at the meeting.

And what happens when several participants bring their own assistants? A single conversation could be captured by multiple tools, each creating its own record under different contractual and technical arrangements. What was intended to be a conversation between a small group of people can quickly become information held across several systems, none of which the meeting organiser necessarily controls or has assessed.

Traditionally, meeting notes captured what mattered rather than every word spoken. AI meeting assistants change that. Depending on their configuration, they can turn an entire conversation into a searchable, shareable and potentially long-lived record. Even information that wasn't relevant to the meeting can be captured, retained and used long after the conversation has ended.

Under New Zealand's Privacy Act 2020, organisations must only collect personal information where it is necessary for a lawful purpose connected with their functions or activities. There's a difference between needing a record of decisions made during a meeting and needing a word-for-word transcript of everything said. Yet increasingly, technology encourages us to collect everything simply because we can.

An external participant's AI assistant adds another layer of complexity. Even if your organisation has decided that recording a particular meeting isn't necessary or appropriate, someone else may have made a different decision and brought their assistant along.

Many people assume that because an AI assistant announces its presence, the privacy obligations have somehow been addressed. But knowing a meeting is being transcribed isn't the same as understanding what will happen to the information afterwards. A recording notification alone may not satisfy the Privacy Act's requirements to inform individuals about the collection of their personal information.

The practical challenges can become even more complicated after the meeting. How do you apply retention rules to a transcript held in someone else's system? What happens if someone requests access to information about them that was captured during the meeting? And how do you ensure confidential or personal information isn't subsequently shared more widely than anyone intended?

There's also the question of accuracy. AI-generated summaries can be useful, but they're not always reliable. Comments can be attributed to the wrong person; context can be lost and tentative suggestions presented as agreed decisions. Two people could leave the same meeting with different AI-generated accounts of what happened, each potentially being relied upon as an accurate record.

There's a subtler issue too. As recording becomes the default, the nature of conversations may begin to change. People might become more cautious about what they say, less willing to challenge ideas or reluctant to discuss sensitive matters openly. A meeting that once allowed people to explore issues informally can start to feel like a permanent record of every contribution.

Privacy isn't just about preventing information from being lost or disclosed. It's also about creating an environment where people feel comfortable communicating, questioning and sharing information appropriately.

AI meeting assistants offer genuine benefits, but their use should be a deliberate choice rather than something we simply accept because the technology is available. For organisations, that means thinking beyond their own approved tools. Having policies, privacy assessments and safeguards in place is one thing, but what happens when someone outside the organisation brings their own assistant into a meeting?

Organisations should consider setting clear expectations about when AI assistants can be used, particularly where personal or confidential information is being discussed. Employees should also feel comfortable questioning an unfamiliar assistant or asking for it to be removed, even partway through a meeting.

Not so long ago, recording a meeting would usually prompt a conversation about whether it was necessary and whether everyone was comfortable with it. Now an AI assistant can join and begin capturing information with barely a second thought.

Perhaps we've become so accustomed to the convenience that we've stopped questioning the implications.

Once a conversation has been recorded, transcribed and stored in someone else's system, the opportunity to decide whether that information should have been collected has already passed. And controlling what happens next may be difficult.

The most useful question we can ask when an unfamiliar assistant appears in our next meeting may also be the simplest.

“Who's listening, and what are they going to do with what they hear?”