AI Notetakers Move Beyond Meetings, Expanding Workplace Recording Risks

| 2 Min Read
Wearable recorders make meeting policies harder to enforce. Companies need clear disclosure, retention rules and human review for AI-generated records.

AI notetaker policies built around meeting software are already behind the technology. A recording no longer needs a bot in a video call or an alert on a laptop screen: it can start with a tap on a watch. For companies, that shift turns a familiar question—whether to admit a notetaker to a meeting—into a wider problem of recording in offices, hallways and social settings.

The tools meet a real business need. Teams supporting customers can use searchable notes to follow up accurately, and new account managers can learn from a history of customer conversations. AI summaries can help employees keep up with meetings and give colleagues useful context. The value is tangible. So is the risk: recordings can capture sensitive discussions without clear notice, and automated summaries can become an authoritative record despite errors or manipulation.

The central change is physical. Notetaking is moving from software that visibly joins a call to devices worn on the body. Company rules, security reviews and expectations for consent have not kept pace.

Recording has left the video call

On July 28, Granola launched an Apple Watch app. A single tap lets the watch record the surrounding conversation. There is no meeting bot to admit, no participant list to inspect and no laptop screen announcing that a recording is underway. The person wearing the device can capture a conversation without others knowing.

That makes the Apple Watch more than a new place to access meeting notes. It changes who can record, where recording can happen and how easily colleagues can detect it. A policy that applies only when employees use approved video conferencing software misses conversations captured on personal devices or wearable apps.

Events that week highlighted the tension. DEF CON banned Meta-style smart glasses from its conference floor because of recording concerns. Meta had acquired AI pendant company Limitless, bringing always-on personal recording into a major consumer hardware pipeline. A public company has also begun screening attendees for smart glasses before all-hands meetings where unreleased plans are discussed.

The authors describe testing the watch themselves: hidden under a jacket sleeve, it recorded an evening walk and produced a clean transcript. The test points to a practical weakness in policies built around visible recording indicators. A rule can require disclosure, but an organization also has to consider how it will respond when the recording device is difficult to spot.

Recording rules vary by state in the United States. California requires consent from everyone involved in a confidential conversation, while many other states permit recording with only one party’s consent. But a legal rule does not create a workplace norm. Employees and visitors still need to know what the organization expects when someone turns on a recorder.

The transcript can be steered—or simply wrong

Even when everyone knows a notetaker is running, its output is not a neutral account. People have described using prompt injection during live meetings, narrating preferred outcomes to a notetaker before others arrive, and crashing competitors’ tools to prevent a record from being produced. The techniques vary, but the common point is that participants can try to shape what the system preserves.

Errors need no deliberate interference. A transcription model can mishear a word and then interpret the rest of the conversation in a way that fits the mistaken version. People with accents the system handles poorly may face more of this risk. When an AI summary assigns decisions and action items, a small transcription error can affect what colleagues believe was agreed.

Some notetakers also pull in information from earlier meetings and background on participants. That context can make notes more useful, but it complicates sharing. A summary sent to an outside party may reveal reasoning or details from conversations that party never attended. Organizations need to consider what the system includes, who can access it and whether an apparently routine recap carries context beyond the meeting itself.

A case involving Campbell’s shows how recording risk extends beyond AI tools. Last November, the company’s security executive was terminated after an employee secretly recorded a meeting that the executive believed was a private lunch conversation. The recording surfaced months later through an employment lawyer, and the executive was gone within days. The example does not involve a meeting bot. It shows that a single person with a device can create a lasting record of a conversation others thought was private.

Policies need to cover people and devices

Organizations asking their CISOs for notetaker policies should start with a clear disclosure rule. If recording is underway, everyone present should know. One suggested approach is to add “+ AI notetaker” to the recorder’s display name in a video meeting. For in-person settings, companies need an equivalent way to make recording visible and to tell people what happens to the output.

The rule should apply to employees as well as outside vendors. Secretly recording a colleague should carry a clear consequence, regardless of whether the device is a phone, laptop, watch or dedicated recorder. A policy that governs only approved software leaves the most discreet recording methods outside its scope.

Meeting categories should be decided in writing. A company might prohibit recording in M&A discussions, legal meetings, HR conversations and executive board sessions. The precise list will depend on the organization, but the decision should be made before a tool’s default settings or an employee’s habits set the practice.

Companies also need rules for other organizations’ devices. An auditor’s laptop may have to meet internal standards before it connects to company systems; recording tools deserve similar attention. Employees should know when an outside party is recording, and the company should establish whether it can receive a copy of the resulting record.

Set rules for data, vendors and review

Retention deserves a deliberate choice. A full transcript and a short summary are different kinds of corporate records, with different levels of detail and exposure. Organizations should decide whether transcripts are deleted by default, subject to legal holds, or retained for a defined period. Leaving the decision to a vendor’s default setting hands control of a sensitive record to a product configuration.

Vendor review matters because these companies may hold recordings of confidential customer, employee and business discussions in their cloud services. Security controls, data reuse and contract terms should be examined before employees adopt the tools. The source points to the class action Brewer v. Otter.ai, which alleges that Otter’s notetaker recorded meetings without everyone’s consent and used recordings to train models, as well as the sudden shutdown of Novacy. Together, those examples underscore why organizations need to understand how recordings are handled and what happens if a provider’s service changes or ends.

Finally, any transcript treated as an official record needs a named human owner. That person should check whether the notes accurately capture the discussion, rather than assuming the model’s output is reliable because it reads cleanly. Employees may be tempted to steer the system toward preferred conclusions. Reviewing a transcript for accuracy is responsible; quietly shaping the record to serve an organizational agenda can create a deeper credibility problem.

The practical shift is already clear: recording can follow an employee into any conversation where a wearable device is present. Policies should define disclosure, restricted meeting types, retention, vendor handling and review in terms that apply across devices. The microphone may be in the room before anyone notices it; the organization’s rules need to be ready before the transcript becomes the record.

Source: John Brown · www.csoonline.com

Comments

Please sign in to comment.