The Chatbot Wiretap Wave: Why Every Chatbot Wiretapping Lawsuit Targets the Deployer, Not the Vendor

A 1967 California penal statute is now the basis for the fastest-growing category of AI litigation in the country. If you run a customer-facing chatbot and a third-party vendor touches the transcript, you may already be a defendant-in-waiting. The chatbot wiretapping lawsuit is not a theoretical risk. An analysis of 284 deployer-facing AI litigation matters found these filings grew from 2 in 2021 to 30 in 2025. More were filed in that single year than in the prior four combined. And the exposure per violation is $5,000 in statutory damages, no proof of harm required.

Key Takeaways

• Wiretap claims under the California Invasion of Privacy Act (CIPA) target the company that deploys the chatbot on its site, not the vendor that built it. You cannot outsource this risk through your vendor's terms of service.
• Complying with California's chatbot-disclosure law (SB 243) does not protect you from a separate wiretap class action. Disclosure and interception consent are legally distinct obligations.
• Courts are split on whether a third-party chat vendor is a passive "recorder" (no CIPA violation) or an independent "eavesdropper" (violation). The outcome depends on your technical architecture, specifically on whether a non-consented third party has access to raw message content in transit.
• The defendant pool is expanding beyond retail and tech into healthcare, insurance, dental practices, and universities. Nearly 3,968 CIPA cases have been tracked in California alone as of mid-2026.
• The strongest defense is architectural: design your data flows so no unconsented third party ever touches raw conversation content. A privacy policy revision alone will not save you.

What Is CIPA, and Why Does It Apply to Chatbots?

CIPA is California Penal Code § 631(a), originally written to prevent telephone wiretapping. It prohibits the intentional interception of communications "in transit" by a third party without all-party consent. The statute predates the internet by decades. It does not mention chatbots, cookies, or pixels. It does not need to.

Plaintiffs' attorneys have mapped its language onto a simple fact pattern: a user types a message into a chat widget on your website, and that message is routed through (or stored by) a third-party vendor the user never consented to interact with. Under the plaintiffs' theory, the vendor is an eavesdropper on a two-party conversation between the user and your company. 83% of all such claims are brought under § 631(a).

The statute's damages provision is what makes this commercially dangerous. Section 637.2 provides $5,000 per violation, and "per violation" can mean per message, per session, or per user, depending on how the court construes it. No actual harm needs to be shown. Multiply that by a class of website visitors and the numbers get large quickly. One tracker reports over $34.65 million in publicly disclosed CIPA settlements across roughly 67 cases as of 2026.

Why Does the Deployer Get Sued Instead of the Vendor?

Because the deployer is the one with the relationship to the user. The wiretap theory targets the company that puts the chatbot on its own website, not the company that built the underlying technology. Your vendor's terms of service might indemnify you, or they might not. Either way, you are the named defendant.

This is the part most engineering and product leads miss. The decision to embed a third-party chat widget is typically made by a product manager or a marketing team. The legal exposure lands on the company. The vendor is, at most, a co-defendant or a cross-claim target after the fact. Your procurement process for chat tooling is, functionally, a litigation-risk decision.

What Does the Case Law Actually Say?

It says different things depending on which courtroom you are in. The law is genuinely unsettled.

Cases That Survived Dismissal

In August 2025, a federal court in the Northern District of California denied a motion to dismiss a CIPA class action against ConverseNow, an AI customer-service vendor that powers restaurant phone and text ordering. A plaintiff alleged her call to a Domino's location using the service was intercepted and recorded without consent. The court let the claim proceed.

In Jones v. Peloton, a California district court allowed CIPA claims past the pleading stage because the plaintiff described in detail how a third-party chatbot allegedly intercepted communications. The court treated the vendor itself as a potential "eavesdropper," not merely a tool the site owner directed.

Cases That Were Dismissed

In February 2026, a Los Angeles Superior Court judge granted summary judgment for an online retailer, finding the plaintiff could not show that a third-party chat vendor actually read or attempted to read message contents in transit. CIPA requires interception, not just logging.

The Ninth Circuit, in an unpublished decision in Gutierrez v. Converse, dismissed CIPA claims after finding no evidence of interception "while in transit" and declined to decide whether CIPA applies to website chats at all. Unpublished means non-precedential. It tells you where the wind blows without binding anyone.

What Determines the Outcome?

The technical facts. Specifically: does a third party that the user did not consent to have real-time or near-real-time access to raw message content during the conversation? If yes, courts are more likely to treat that party as an eavesdropper. If the vendor functions as a tool under the deployer's control, with no independent access to content, courts lean toward dismissal.

This is not a legal distinction. It is an engineering distinction that courts are being asked to evaluate.

Does Complying with SB 243 Protect You from Wiretap Claims?

No. This is the most counterintuitive gap in the compliance landscape right now. A company that fully satisfies California's chatbot-disclosure law (SB 243) can still face a separate wiretap class action if its chatbot records conversations without ECPA-compliant consent. The two obligations are legally distinct.

SB 243 requires you to disclose that the user is interacting with a bot. CIPA requires you to obtain consent before a third party intercepts the communication. Disclosure is not consent. Telling someone "you are chatting with a bot" is not the same as telling them "a third-party vendor will receive and process the contents of this conversation." You need both, and one does not satisfy the other.

If you have a legal team that checked the SB 243 box and moved on, go back and ask them about interception consent specifically. These are parallel tracks.

Which Industries Are Being Targeted?

All of them, increasingly. The initial wave hit retail and tech companies with high-traffic customer service chat widgets. That wave has broadened. 2025 defendants included healthcare companies, insurance providers, dental practices, and universities.

Litigation tracker data as of mid-2026 shows nearly 3,968 tracked CIPA cases in California, 811 in Florida, and 108 in Illinois. Retail is the most-targeted industry at 1,817 cases, followed by technology (542) and professional/technical services (447).

If you operate a website with a chat feature and your customers include California residents, you are within the statute's reach regardless of where you are incorporated.

What About AI Training Data as a Wiretap Theory?

This is the newer and potentially more expansive theory. Some suits now allege that user chat data is being accessed by chat-tool providers to improve their services or train AI models, broadening the wiretap theory beyond simple interception into downstream use of conversation content.

In May 2026, a proposed class action was filed against OpenAI in the Southern District of California alleging that Meta Pixel and Google Analytics were embedded in the ChatGPT.com site, intercepting user queries and leaking them to advertisers. This extends the theory from "your chatbot vendor eavesdropped" to "your analytics stack leaked the conversation to a fourth party."

If this theory gains traction, the attack surface expands beyond the chat vendor itself to every piece of JavaScript on the page where the chat widget loads. Your tag manager becomes a liability vector.

What Does a Defensible Architecture Look Like?

The legal question maps onto a concrete engineering question: at each point in your chatbot's data flow, who has technical access to raw conversation content, and did the user consent to each of those parties?

Here is what that looks like in practice:

1. Map every party that touches transcript data. Your chat widget provider. Your analytics tools. Any pixel or tag on the page where the widget loads. Your logging and monitoring stack. Your AI inference provider, if the chatbot routes messages to a model API. List them all.
2. For each party, determine whether they have access to raw message content. Some analytics tools capture form inputs by default. Some session-replay tools record keystrokes. If any of these are running on the same page as your chat widget, they may be intercepting conversation content without your knowledge.
3. For each party with content access, determine whether you have user consent. Not disclosure. Consent. Obtained before the conversation begins, not buried in a privacy policy footer.
4. Evaluate whether any party qualifies as an "eavesdropper" under the legal test. The key question courts are asking: does this vendor operate as a tool under your direction (a "recorder"), or does it independently access and process the content for its own purposes (an "eavesdropper")? If the vendor uses conversation data to train its own models, improve its own product, or share with its own partners, you are on the wrong side of that line.
5. Architect to minimize third-party content access. If you can process conversations through a first-party system and only pass necessary data (not raw transcripts) to third parties, you reduce your surface area. Encrypt content at rest. Limit who can decrypt it. Document the data flow.

This is not a legal opinion. It is an engineering checklist derived from reading the complaints and the rulings. Your counsel should review it. But your counsel cannot build it. That part falls on your engineering team.

How Is the Legislative Landscape Shifting?

In the first weeks of 2026, 78 chatbot-related bills were filed across 27 states. At least six states advanced their own chatbot legislation past committee or a full chamber in the two months after California's companion-chatbot law took effect on January 1, 2026.

On the defense side, California's Senate Bill 690 aims to curb what critics call "shakedown" CIPA litigation by clarifying the statute's scope relative to ordinary website tracking. Whether that passes, and whether it retroactively affects pending claims, is open.

The volume of legislative activity tells you two things. First, legislators recognize that a 1967 wiretapping statute is being stretched to cover interactions its authors never imagined. Second, they have not yet agreed on how to fix that. Until they do, the existing statute and its $5,000-per-violation damages provision remain the operative law.

What Should You Do This Week?

If you are an engineering, product, or legal lead at a company running a customer-facing chatbot, here are the concrete steps that map to the current state of the case law:

• Audit your chat widget's third-party dependencies. Run your chat page through a network inspector. Identify every domain that receives data when a user types a message. If any of those domains belong to parties not disclosed in your consent flow, you have a gap.
• Separate your disclosure compliance from your consent compliance. Verify that your SB 243 disclosure (if applicable) exists, then verify separately that you have interception consent for every third party in the data flow. Two checklists, not one.
• Review your chat vendor's data-use terms. Does the vendor retain conversation content? Use it for model training? Share it with sub-processors? If yes, the vendor may be classified as an eavesdropper rather than a recorder under current CIPA case law.
• Remove unnecessary analytics from chat pages. If your session-replay tool or marketing pixel does not need to run on the page where users interact with your chatbot, remove it from that page. Every additional script is an additional party with potential content access.
• Document your architecture decisions. If you do face a claim, the cases that were dismissed turned on the defendant's ability to show that no third party independently accessed content in transit. That showing requires technical documentation. Data flow diagrams. Access control records. Retention policies. Build the record now.

Why Architecture Is the Only Durable Defense

Privacy policies can be rewritten. Consent banners can be redesigned. But the underlying question in every chatbot wiretapping case is a question about plumbing: who actually touches the data, and when?

A consent banner that says "we use cookies" does not help if a chat vendor is independently processing full conversation transcripts. A privacy policy that lists twelve categories of data recipients does not help if users click "I agree" without reading it and a court decides that does not constitute meaningful consent to interception.

The companies that have won dismissals did so by demonstrating, with technical evidence, that no third party intercepted content in transit. The companies that lost motions to dismiss could not make that showing. The distinction was not in their legal arguments. It was in their system design.

This is where the calculus favors building privacy into the product rather than bolting it onto the policy. If your chatbot architecture routes messages through a first-party system, encrypts content at rest, and limits third-party access to what is strictly necessary (with consent obtained for each accessor), you have a defensible position. If your architecture mirrors every message to a vendor who uses it for their own purposes, you have a plaintiff's exhibit.

We built Selina with this principle. Conversations are processed through a stack of frontier models, routed per task via API. Memory is encrypted at rest. We are not injecting analytics pixels into your conversation. We are not training on your data. That is an architecture choice, not a marketing claim.

The wave of chatbot wiretap litigation is not going to recede before the law catches up with the technology. Until it does, your best defense is the one you build into the system before anyone sends a demand letter.

If you want to see what a privacy-first AI assistant looks like from the user side: Start a free 7-day trial, no card required.

Michael C.

Founder & Principal Engineer, Selina Labs

Michael builds Selina, a privacy-first AI that remembers you across conversations. He ships security-sensitive AI in production — real attacks, real fixes, measured in minutes and dollars — and writes about privacy, security, and LLMs from that seat. Top Rated Plus and expert-verified on Upwork.

• www.upwork.com
• x.com
• medium.com
• selinalabs.io