Why Letting an AI Notetaker Join Interviews and HR Investigations Destroys Your Best Evidence

How Vendor Model Training Turns a "Note-Taker" Into a Third-Party Wiretapper

When an AI transcription vendor retains call audio or transcripts to train its own machine-learning models, federal courts refuse to treat the bot like an employer's passive notepad. Under In re Otter.ai Privacy Litigation (N.D. Cal. Aug. 13, 2026), commercial model training plausibly converts the vendor into an independent third-party eavesdropper under California wiretap law.

Join almost any remote hiring loop or employee-relations intake call and an uninvited tile quietly enters the grid: "Hiring Manager's Notetaker." Recruiters and investigators defend the habit as harmless administrative hygiene because nobody wants to miss a quote while holding eye contact. That convenience argument rests on a shaky legal premise. It assumes a cloud AI vendor occupies the exact same statutory seat as a yellow legal pad on your desk.

On August 13, 2026, Judge Eumi K. Lee rejected that premise at the pleading stage in In re Otter.ai Privacy Litigation (Brewer v. Otter.ai, Inc.), Case No. 5:25-cv-06911-EKL (N.D. Cal.). The court declined to dismiss core claims under the Electronic Communications Privacy Act (ECPA), the California Invasion of Privacy Act (CIPA, Cal. Penal Code § 631(a)), and the Illinois Biometric Information Privacy Act. A vendor that strictly stores encrypted audio for one customer's sole use acts like a passive tape recorder under the party exception. Once that vendor ingests speech or transcripts to train its own machine-learning models, it plausibly acts as an independent third-party wiretapper reading conversations in transit for its own gain.

For multi-state people operations teams, the statutory exposure compounds across every remote dial-in. California CIPA authorizes $5,000 per violation in statutory civil damages under Cal. Penal Code § 637.2(a)(1) even without financial loss. All-party wiretap statutes also protect participants calling from Pennsylvania (18 Pa. C.S. § 5704(4)), Florida (Fla. Stat. § 934.03), Illinois (720 ILCS 5/14-2), Massachusetts (M.G.L. c. 272, § 99), Washington (RCW 9.73.030), and Maryland (Md. Code, Cts. & Jud. Proc. § 10-402).

The sharpest trap for Employee Relations sits inside Cal. Penal Code § 632(d). Aside from prosecuting the illegal recording itself, no evidence obtained by eavesdropping upon or recording a confidential communication without all-party consent is admissible in any judicial, administrative, or other proceeding. Record a misconduct respondent's admission through a manager's unannounced personal AI bot, fire for cause, and watch the trial judge bar your primary investigation record from court.

Speaker Diarization Under Illinois BIPA and Video Interviews Under AIVIA

Clicking past a generic video-platform banner never satisfies biometric statutes when speaker-identification tools map vocal geometry across calls under 740 ILCS 14/15. Illinois employers also face separate pre-interview notice, prior consent, sharing, and 30-day deletion mandates under 820 ILCS 42 whenever AI evaluates applicant video interviews.

A passive pop-up chime on Zoom or Teams does not cure biometric liability. Commercial AI meeting assistants pitch "speaker diarization"—separating Speaker A from Speaker B and matching recurring voices to calendar profiles. Under the Illinois Biometric Information Privacy Act (740 ILCS 14/10 and 14/15), extracting vocal features to identify speakers plausibly captures a regulated voiceprint, one of the exact claims sustained in Otter.ai.

Under 740 ILCS 14/15(a) and 15(b), an employer cannot capture or obtain a voiceprint without first publishing a public retention-and-destruction schedule and securing an informed written release before capture begins. Verbal assent at the start of an interview falls short of BIPA's text. Even after Illinois P.A. 103-0769 (SB 2979, effective Aug. 2, 2024) shifted BIPA recovery from per-scan to per-person calculation, 740 ILCS 14/20 imposes $1,000 per aggrieved person for negligent violations and $5,000 per aggrieved person for reckless or intentional violations, plus attorneys' fees. Run 300 Illinois engineering candidates through an unconsented personal bot and the statutory floor hits six or seven figures fast.

Teams governing AI and work tools in recruiting must also apply the Illinois Artificial Intelligence Video Interview Act (820 ILCS 42/1 et seq.) whenever AI analyzes or evaluates applicant video interviews for positions based in Illinois. Employers must notify applicants before the interview, explain how the AI works and what general characteristics it evaluates, obtain prior applicant consent, limit sharing to evaluators who need access, and delete all copies within 30 days of an applicant's request under 820 ILCS 42/15.

Why Raw AI Transcripts Sabotage HR Investigations and Upjohn Privilege

Even with signed consent forms, automated speech-to-text bots manufacture discoverable impeachment traps in workplace investigations by dropping negations and misattributing crosstalk. Routing counsel-directed interviews through consumer SaaS tools also threatens attorney-client privilege and work-product protection under Upjohn.

Assume IT locks down a zero-training enterprise license and every employee signs a release. Should Employee Relations bring an AI transcription bot into a harassment, retaliation, or whistleblower investigation anyway? Almost never.

Large speech-to-text models optimize for conversational fluency rather than courtroom phonetic precision. Under emotional tension, accents, or overlapping speech, automated transcription regularly drops one unstressed syllable—turning "I didn't cut her territory after her leave" into "I did cut her territory after her leave"—or assigns a witness's heated remark to the HR investigator. When the investigator corrects that draft two days later, plaintiff's counsel demands both the raw machine transcript and the edit history in discovery, framing ordinary cleanup as witness tampering—the exact credibility collapse detailed in our review of conflicted HR investigations and major jury verdicts.

Privilege makes automated bots riskier still. When internal or outside counsel directs a confidential workplace investigation under Upjohn Co. v. United States, 449 U.S. 383 (1981), counsel's structured interview memorandum weaves factual synthesis together with legal analysis and enjoys strong work-product protection under Federal Rule of Civil Procedure 26(b)(3). Raw verbatim bot transcripts contain zero attorney mental impressions, making them easier to compel in litigation, while routing privileged interviews through a consumer SaaS plugin that allows vendor review or model training risks waiving privilege completely.

Call-by-Call Governance Matrix and Pre-Interview Consent Protocol

Enforce explicit meeting-bot tiers at your Zoom, Microsoft Teams, or Google Meet admin boundary instead of policing individual browser extensions by memo. Pair approved recruiting use with prior written releases and block bots completely on ER and privileged calls.

=== PART A: PRE-INTERVIEW WRITTEN AI NOTETAKER & BIPA/AIVIA RELEASE ===
[Include in ATS Interview Scheduling Form — Signed Prior to Call]
1. Zero-Training Enterprise Scope: With your optional consent, [Company] uses
   [Approved Vendor] solely to draft text notes of structured interviews for
   the hiring team. Under [Company]'s enterprise contract, your audio and
   transcript are NEVER used to train external AI models.
2. Biometric & AI Disclosure (740 ILCS 14/15 & 820 ILCS 42): Persistent
   voiceprint indexing is disabled. Any transient vocal data and interview
   records are permanently destroyed within 30 calendar days of the call, or
   within 30 days of your deletion request to [Email] per 820 ILCS 42/15.
   No AI tool scores facial expressions or makes hiring decisions.
3. Penalty-Free Opt-Out: Declining transcription never affects your candidacy.
   [ ] I CONSENT to enterprise transcription under the schedule above.
   [ ] I DECLINE transcription; please take manual written notes only.
   Applicant Signature: ________________________. Date: ______________.

=== PART B: LIVE 15-SECOND CALL-OPENING CHECK (INTERVIEWER) ===
"Before we start, you chose [Consent / Manual Notes] on your scheduling form.
Confirming nobody has a personal AI bot running, and that you remain comfortable
with [our enterprise note tool / manual written notes] for our call today?"

Enforcing Tenant-Level Controls Across People Ops, IT, and Legal

Handbook policies collapse the moment a hiring manager installs a free Chrome recorder or clicks "Admit" on an unknown waiting-room bot. Enforce technical blocks at the video-platform tenant boundary and divide ownership cleanly across three desks.