Chatbot Wiretapping Lawsuits: The Third-Party Theory in Action

    Updated 2026-06-29 · 5 cases cited · 6 outbound resources

    Chatbot wiretapping suits are the fastest-growing category of CIPA litigation. Plaintiffs argue that when a website embeds a third-party chat tool — Drift, Salesforce, LivePerson, Intercom, Zendesk, and others — the vendor silently intercepts every message the visitor types, violating CIPA § 631(a)'s prohibition on third-party interception.

    TL;DR

    • —Hundreds of chatbot-CIPA cases were filed in 2023–2026, with Drift and Salesforce as the most frequently named vendors.
    • —The legal theory is § 631(a)(2) — the vendor reads communications in transit without consent of all parties.
    • —Courts have repeatedly rejected the 'party exception' defense for SaaS vendors processing the chat.
    • —AI chatbots add a § 632 dimension because LLM training on chat logs may constitute additional recording.

    Got a demand letter? Start by checking your Shopify or ecommerce page after Reject All. For checkout testing and help fixing consent and pixel issues, explore got pixels?.

    Why chatbots are uniquely exposed

    Unlike a tracking pixel that captures behavioral metadata, a chatbot intercepts the literal contents of a conversation — exactly the scenario CIPA was written to prevent. Plaintiffs do not need to argue about whether 'clicks' qualify as a communication; the user is typing sentences.

    Most chatbot vendors operate a SaaS model where the chat session is processed on the vendor's servers, transcribed, stored, and often fed back into product analytics or AI training. Every step of that pipeline is a potential CIPA violation if the visitor was not given an opt-in choice before the chat opened. The theory is not confined to California either: federal ECPA and state analogues like WESCA support the same claims, and our guide to the wiretap statutes behind pixel litigation shows how they overlap.

    The leading vendors named in suits

    Drift (acquired by Salesforce in 2024) is named in the largest number of complaints. Salesforce's broader Marketing Cloud and Service Cloud chat products are next, followed by LivePerson, Intercom, Zendesk, Glassix, and Ada. AI-native vendors (Forethought, Cresta, Decagon) have begun appearing in late-2025 filings.

    Operators using these vendors are sued as primary defendants; the vendors are typically added later for indemnification reasons.

    The 'party exception' defense — and why it fails

    Defendants routinely argue that the vendor is not a third party because it acts as the website operator's 'extension' or 'agent.' Courts have rejected this in Licea v. Caraway Home (C.D. Cal. 2023), Byars v. Hot Topic (C.D. Cal. 2023), and several follow-on rulings. The reasoning: the vendor processes the data for its own commercial purposes (analytics, training, product improvement), making it an independent party.

    The leading contrary authority (Graham v. Noom) is a district court decision in a different circuit and has been distinguished by California courts.

    The AI chatbot wrinkle

    When the chatbot is powered by an LLM, two new CIPA exposures emerge. First, the chat transcript is often retained for fine-tuning or RLHF, which courts may treat as a separate § 632 'recording' violation. Second, the LLM provider (OpenAI, Anthropic, Google) is a fourth party in the data flow, multiplying the interception count.

    Late-2025 complaints have begun to plead these theories explicitly, and at least one ruling has allowed the LLM-vendor claim to proceed past the pleading stage.

    Got a demand letter? Start by checking your Shopify or ecommerce page after Reject All. For checkout testing and help fixing consent and pixel issues, explore got pixels?.

    What plaintiffs typically allege

    A typical chatbot complaint pleads (a) the visitor opened the chat without seeing an opt-in disclosure, (b) the visitor typed at least one message, (c) that message was transmitted to the vendor's servers in real time, (d) the vendor processed the message for its own commercial benefit, and (e) the visitor was located in California. The complaint usually attaches screenshots and HAR-file evidence captured from a test session.

    How to reduce exposure

    1. Gate chatbot loading behind opt-in consent (Consent Mode or equivalent). 2. Disclose the vendor and the data flow before the first message can be sent — an 'open' state with a visible notice and an explicit Start button is the safest pattern. 3. Sign a DPA that prohibits the vendor from using chat data for its own purposes, then enforce it. 4. Disable training and analytics retention by configuration where the vendor permits. 5. For AI bots, prefer self-hosted or zero-data-retention LLM tiers.

    Key cases

    Licea v. Caraway Home, Inc.

    C.D. Cal. 2023, No. 22-cv-1791

    Denied motion to dismiss CIPA claim against a website using a third-party chat. Court rejected the 'party exception' defense and found the vendor to be a separate party for § 631(a) purposes. Frequently cited in subsequent chatbot rulings.

    Byars v. Hot Topic, Inc.

    C.D. Cal. 2023, No. 22-cv-1652

    Allowed CIPA chatbot claim to proceed where the chat vendor recorded conversations for its own analytics. Influential application of Javier to chat tools.

    Valenzuela v. Keurig Green Mountain, Inc.

    N.D. Cal. 2023, 674 F. Supp. 3d 751

    Sustained CIPA claim against Keurig for embedding a third-party chat tool that captured visitor messages. The court emphasized that the chat opened automatically without consent.

    Saleh v. Nike, Inc.

    C.D. Cal. 2023, No. 22-cv-08989

    Class action against Nike for use of a Salesforce-powered chat. Settled in 2024 for an undisclosed amount.

    Graham v. Noom, Inc.

    N.D. Cal. 2021, 533 F. Supp. 3d 823

    Outlier ruling that adopted the 'party exception' to dismiss a chatbot CIPA claim. Distinguished or rejected by most subsequent California rulings.

    FAQ

    Is a 'chat opened' badge enough disclosure?

    Almost certainly not. Disclosure must precede transmission of any communication, must identify the vendor and the data flow, and ideally must require an affirmative click before the user can type.

    What if the chat is hosted on the operator's own infrastructure?

    Self-hosted chat removes the 'third-party interception' theory but does not eliminate § 632 recording exposure. You still need consent before recording the conversation.

    Does the vendor's DPA protect me?

    A DPA shifts contractual risk but does not create consent under CIPA. The visitor — not the vendor — is the party whose consent matters. A DPA also does not protect the vendor itself from being added as a defendant.

    Are AI chatbots higher risk than human chat?

    Generally yes. AI chatbots involve more data flows (the LLM provider), longer retention windows (training data), and more sophisticated analytics, all of which expand the universe of CIPA-relevant interceptions.

    Further reading

    Related reading

    Understand the tracking behind your demand letter

    Start with a free Reject All scan of your Shopify or ecommerce page. Enter your email to unlock detailed findings and reports for your team and lawyer. got pixels? goes deeper into checkout and other journeys, helps with remediation, and lets you retest changes. A scan cannot confirm or dismiss a legal claim.

    This article is general information about California Invasion of Privacy Act litigation, not legal advice. Case citations are drawn from public dockets and law-firm commentaries; verify before relying on them. Consult qualified counsel for advice on your specific situation.