California CIPA and AI Sales Calls: The 2026 Operator Guide
California is the highest-risk jurisdiction in the US for AI call recording and AI listening. CIPA (Cal Penal Code 631-632) requires all-party consent, and the active Ambriz v. Google interception theory extends exposure to AI listeners that do not even record. Here is what sales teams need to know.
Last verified May 2026. This page is informational only; consult qualified California counsel for your specific deployment.
§The CIPA Statute in Plain English
The California Invasion of Privacy Act sits at Cal Penal Code Sections 630 to 638. The two sections that drive AI call-recording risk are:
Section 631: Wiretapping (interception of communications)
Prohibits intentional intercept, eavesdrop, or wilful disclosure of confidential communications without consent of all parties. The Ambriz theory leans on this section to argue that AI processing of call audio constitutes interception regardless of whether traditional recording occurs.
Section 632: Recording of confidential communications
Prohibits recording of confidential communications without consent of all parties. This is the section that catches Fathom, Fireflies, Otter, Gong, Chorus, and Avoma if call recording happens without disclosure to all participants. The conservative reading is that all participants must affirmatively consent or have constructive notice; mere participation is not consent.
Section 637.2 provides the private right of action: $5,000 per violation or three times actual damages. This is the section that funds the plaintiffs' bar class actions and drives the bulk of enforcement risk. Section 638 covers unauthorised connection to telephone lines but is less commonly invoked in AI-call cases.
§The Ambriz v. Google Interception Theory
Ambriz et al. v. Google LLC, filed in 2024 in the Northern District of California, alleges that Google Meet's AI summary, transcription, and caption features constitute interception of confidential communications under CIPA Section 631 even when no traditional recording is created. The theory is novel and consequential.
Traditional CIPA interpretation focused on recording (the physical or digital capture of audio for replay). The Ambriz theory argues that any AI system that listens to call audio for processing, whether or not the audio is persisted, intercepts the communication and triggers CIPA Section 631. If accepted by courts, this expands CIPA exposure from recording-tool vendors to any AI-listener vendor including Microsoft Teams Copilot, Zoom AI Companion, Webex Audio Intelligence, plus all sales-side vendors (Gong, Chorus, Clari Copilot, Avoma, Fathom, Fireflies).
As of May 2026 the Ambriz case has not been decided on the merits. Settlements in adjacent cases (Williams v. DDR Media, Calhoun v. Cookies SF) signal that defendants are taking the theory seriously and settling rather than litigating to final judgment. The conservative compliance posture for AI sales tools is to operate as if the Ambriz theory will prevail.
§Compliance Posture Recommendations
Based on the conservative reading of CIPA plus the Ambriz theory, sales teams calling California prospects with AI-equipped tools should adopt the following posture. None of this is legal advice; confirm with qualified counsel.
1. Explicit AI disclosure in the call opener
Replace "this call may be recorded" with "this call is handled by an AI assistant and may be recorded; if you do not consent, please disconnect now or ask for a human agent." The explicit AI mention plus the disconnect/opt-out option is the conservative compliant posture.
2. Affirmative consent capture for outbound voice AI
For outbound AI voice calls (Vapi, Retell, Bland, Synthflow deployed for outbound), prior express written consent is required not just for CIPA but for TCPA per FCC Ruling 24-17. The CIPA layer adds requirement that the consent specifically mention AI processing, not just recording.
3. Geographic-routing detection at call setup
Detect California area codes (and resident self-disclosure during the call) at call setup. Apply the explicit AI disclosure plus opt-out for any call where any party is California. Some teams apply the disclosure universally to avoid the per-call routing complexity; this is the safest posture.
4. AI processing of recording must be disclosed even when call is recorded with consent
Consent to recording is not automatically consent to AI processing of the recording under the Ambriz theory. Disclose AI processing of stored recordings (call summaries, AI scorecards, AI search) in privacy notices and in the recording disclosure itself.
5. Audit trail of consent capture
Maintain a verifiable audit trail of consent (recording of the disclosure plus consent acknowledgement, timestamped, retained for the statute-of-limitations period plus margin, typically 4 years from call). The audit trail is the primary defence against a CIPA class action alleging that consent was not obtained.
§Vendor Posture on CIPA
Major AI sales-tool vendors have varying postures on CIPA disclosure. None of these vendors will guarantee CIPA compliance for the customer; the operator carries the compliance burden. The vendor posture is about what features are available to support compliance.
| Vendor | Auto-disclosure feature | CIPA-specific notes |
|---|---|---|
| Gong | Yes (configurable per-org) | Customisable opening disclosure; consent log available |
| Chorus by ZoomInfo | Yes (default opt-in) | Standard disclosure; consent log via ZoomInfo admin |
| Clari Copilot | Yes (configurable) | Manual configuration of California-specific notice; consent log available |
| Avoma | Yes (default) | Standard call-start disclosure; consent capture optional add-on |
| Fathom | Yes (default) | Mandatory bot announcement at meeting join; consent log via account |
| Fireflies | Yes (default) | Bot announcement at join; consent log via account |
| Otter.ai | Yes (default) | Bot announcement; consent log via account |
| Vapi | Operator configures | Disclosure must be implemented in the agent's opening script |
| Retell AI | Operator configures | Disclosure must be implemented in the agent's opening script |
| Bland AI | Operator configures + templates | Disclosure templates available but operator implements |
§The B2B Call Exception (And Why It Does Not Save You)
A common misconception in sales: "CIPA only applies to consumers; B2B sales calls are exempt." This is not correct. CIPA applies to all confidential communications, not just B2C. The relevant distinction is whether the communication is confidential, not whether the parties are individual consumers or business representatives.
California courts have repeatedly held that a typical business sales call (where the prospect would have a reasonable expectation that the conversation is not being broadcast or recorded) is confidential under CIPA. The fact that both parties are at work, calling business numbers, with business email addresses, does not strip the call of confidential status.
Some California courts have carved out narrower exceptions for calls where the prospect has clear notice that recording is happening (an inbound customer service line with prominent IVR disclosure, for example). These exceptions do not extend to outbound sales calls where the AE initiates contact without prior disclosure.
§Settlement and Litigation Reference
Several CIPA settlements in the last 24 months have established the financial weight of the risk:
- + Calhoun v. Cookies SF (2023): $9.75 million class settlement over website session-replay CIPA claims. While not an AI-call case directly, established the financial template for CIPA class action damages.
- + Williams v. DDR Media (2024): $7.5 million settlement involving session replay and AI-adjacent listener technology. Cited by Ambriz plaintiffs as analogue.
- + Salesloft v. State of California (regulatory, 2024): California AG inquiry into sales call recording practices; not a class action but established regulatory interest in the category.
- + Ambriz v. Google (ongoing, filed 2024): The current bellwether case for AI-listener-as-interception theory.
None of these cases have produced a clear judicial determination on AI-listener-as-interception in the sales-call context. Sales teams should assume the most plaintiff-favourable reading until a higher court resolves the theory.