The Wave Continues To Build

This is the third installment in the Holland & Knight Cybersecurity and Privacy Blog's ongoing series tracking the rapid evolution of privacy class actions targeting generative artificial intelligence (GenAI) tools and technologies. A July 2025 post, "Up Next in Privacy Litigation: Consumer Class Actions Targeting Generative AI Tools," examined how plaintiffs' firms were beginning to apply wiretap statutes to GenAI call center tools – most notably in Ambriz, et al. v. Google LLC, 3:23-cv-05437 (N.D. Cal.) – following the playbook established by earlier waves of litigation involving session replay technology, chatbots and tracking pixels. A May 2026 post, "Recent GenAI Class Actions Build on Early Successes and Break New Ground," reported on a string of early court rulings that largely favored plaintiffs, establishing a viable foothold for wiretap claims against AI vendors and noting the emergence of new legal theories targeting technologies beyond the call center, including ambient AI in healthcare settings and AI-powered features embedded in consumer communications platforms.

Now, two new cases filed in late July 2026 demonstrate that this litigation trend is accelerating and expanding into workplaces and restaurant telephone systems, raising novel questions about non-party class theories and federal statutory claims that organizations deploying AI tools must carefully consider.

The New Cases

In Chamberlain v. Granola, Inc., No. 3:26-cv-07926 (N.D. Cal.), a Florida plaintiff alleges that Granola's AI notetaking application secretly intercepted her Microsoft Teams and/or Zoom communications without her knowledge or consent. The plaintiff claims that, unlike competitor AI notetakers that announce their presence by joining as a visible bot, Granola is designed to operate as a feature invisible to other meeting participants, citing Granola's website, which advertises that "Other people in the room won't know it's there." The complaint advances a novel class theory against a non-party to the communication: The plaintiff never used Granola or agreed to its terms but was an unwitting meeting participant recorded by another attendee's software.

In Thompson v. SoundHound AI, Inc., 26-cv-202181 (Alameda County Superior Court), a California plaintiff alleges that SoundHound's "Smart Ordering" phone AI – deployed at several restaurant chains – secretly intercepted her restaurant phone calls. The complaint further alleges that when a consumer dials a restaurant's published number, the call is routed to SoundHound's servers before any human greeting plays, and SoundHound – not the restaurant – answers and transcribes in real time.

Both Granola and SoundHound AI share the core focus of the Ambriz line of call center cases: AI systems intercepting voice communications without the knowledge or consent of all participants. However, these new cases push the legal envelope in several important ways. Unlike Ambriz, which targeted a tool used within a traditional call center environment and asserted only claims under the California Invasion of Privacy Act (CIPA), Granola invokes CIPA, along with the federal Electronic Communications Privacy Act (ECPA) and California Comprehensive Data Access and Fraud Act (CDAFA), treating the interception as unauthorized computer data access. SoundHound AI relies on CIPA § 632.7, which applies to any recording of a cellular or cordless communication without regard to confidentiality. This is a significantly lower threshold than the confidentiality requirement at issue in the call center cases. Together, these cases expand the theory from consumer-facing transactions into the workplace, direct user plaintiffs to non-party classes and state-only claims to federal statutory theories.

Next Steps

Organizations deploying AI tools that interact with voice communications – whether in customer service, professional collaboration or telephone infrastructure – should assess their use of these technologies and evaluate potential risks and mitigation efforts. This includes assessing indemnification and liability-limiting provisions when onboarding AI vendors, consulting with defense counsel on whether similar tools have been targeted in litigation, ensuring privacy policies accurately disclose AI-related data collection and use, and obtaining and documenting express consent from all participants before AI processing begins. Given the non-party class theory in Granola, organizations should also consider whether their AI tools' default settings expose non-users to interception without consent.

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.