AI Smart Glasses at Work, Part 2: Two-Party Consent and AI Note-Taking Risks Under Wiretapping Laws

The Hidden Legal Minefield: Compliance Concerns with AI Smart Glasses, Part 2 - Two-Party Consent and AI Note-Taking

Smart glasses are moving from novelty to daily tool. Cameras, always-on mics, and real-time AI turn a pair of frames into a recorder, a stenographer, and an analyst you wear on your face.

That mix creates a consent problem. If your device can capture and transcribe nearby conversations without clear notice, you're stepping into wiretap law, privacy torts, and discovery exposure.

The Risk

Recording or transcribing a conversation without permission can violate state wiretapping laws. A number of states require all parties to consent (often cited as 12), including California, Florida, Illinois, Maryland, Massachusetts, Connecticut, Montana, New Hampshire, Pennsylvania, and Washington. Federal law and many states permit one-party consent, but that does not apply in places where people reasonably expect privacy.

AI note-taking changes the posture of risk. Glasses can passively capture speech all day, produce transcripts no one knew existed, and store them in searchable form. If there's no obvious indicator, consent and notice get even harder. That's how accidental crimes happen.

For a practical overview of state recording rules, see the Reporters Committee's guide to consent laws here. California's Invasion of Privacy Act (e.g., Penal Code § 632) is a common flashpoint; statutory text is available from the state legislature here.

Relevant Use Cases

• Sales reps wearing AI glasses that auto-transcribe client meetings without explicit consent from all participants
• Managers using AI note-taking during performance reviews, disciplinary meetings, or interviews
• Medical professionals recording patient consultations for AI-generated documentation
• Employees on phone calls where the other party is in an all-party consent state
• Anyone wearing recording-capable glasses in restrooms, locker rooms, medical facilities, or other high-privacy areas
• Workers transcribing confidential business discussions or trade secret conversations
• OSHA inspectors using AI glasses (expanded deployment announced for 2025) to record workplace inspections without proper protocols

Why It Matters

Violations can bring criminal exposure and civil liability. Tiny LED indicators are easy to miss, so "silent" recording is a real threat vector. AI-generated transcripts created without consent open up consent claims, privacy claims, and separate obligations under data minimization and retention rules.

Discovery risk is real. If employees generate transcripts, those records may be discoverable, create new custodians, and expand the universe of potentially responsive data across vendors and cloud services.

Practical Compliance Playbook

• Set policy boundaries: Define where and when recording-capable glasses are allowed and prohibited (by role, location, and event type).
• Get consent first: Obtain explicit consent from all parties before recording or transcribing. Consent banners on video calls often do not cover wearables.
• Give clear notice: Use visible signage and verbal disclosures. Don't rely on device LEDs that people won't see.
• Use technical limits: Enforce geofencing and auto-disable in prohibited areas (restrooms, medical spaces, legal meetings, secure labs).
• Log activity: Record who activated audio/transcription, when, where, and for how long. Tie logs to user identity.
• Train employees: Provide state-specific briefings, plus travel playbooks for interstate meetings and calls. For practical AI training resources by role, see Complete AI Training.
• Disclose AI note-taking: Tell participants when transcription is on, how it will be used, and how to opt out. Offer a non-recorded channel.
• Apply data minimization: Default to off. Limit what is captured, suppress bystander audio, and redact sensitive content at the edge where possible.
• Govern transcripts: Classify, restrict access, set retention, and enforce deletion. Keep transcripts out of personal storage and unmanaged apps.
• Vendor due diligence: Review model providers and device vendors for storage location, secondary use, security, and subprocessor chains.
• BYOD guardrails: Prohibit personal cloud syncing and consumer AI apps for workplace recordings and transcripts.
• Litigation readiness: Map where transcripts live, set legal holds, and standardize preservation across devices and AI services.
• Incident response: Treat nonconsensual recording as a reportable event; define escalation, remediation, and notification steps.
• Audit and enforce: Periodically test controls, spot-check logs, and discipline policy violations consistently.

A Note on Cross-Border and Interstate Calls

Two-party consent can apply based on the location of any participant. If one person joins from an all-party state, treat the conversation as all-party. When in doubt, get consent from everyone or do not record.

Questions Legal Should Ask Before Deployment

• What models and sensors are active by default? Can we prove audio is off without explicit user action?
• Where do raw audio and transcripts flow (device, app, vendor, model provider)?
• How do we disable capture in protected areas and privileged contexts?
• What is our legal basis for collection, and how is consent recorded and retrievable?
• What retention windows apply by data type, and who can approve exceptions?
• How are bystander voices suppressed or filtered?
• What is our plan for public spaces, events, and unionized worksites?
• How do we handle DSARs, subpoenas, and third-party requests for transcripts?
• Do vendors fine-tune on our audio or transcripts by default?
• What's the fallback process when participants refuse recording?

Bottom Line

AI glasses can be valuable, and shipments reportedly grew 210% in 2024. The legal exposure is growing just as fast. Treat audio capture and AI note-taking as regulated data collection, not convenience features.

With clear policies, strong technical controls, real consent, and disciplined data governance, you can use the tech without stepping on a legal landmine. Go slow, be explicit, and keep the default set to "off."