Gavalas v. Google LLC

Issue: In *Gavalas v. Google LLC*, Google and Alphabet Inc. argue that product liability, wrongful death, and consumer protection claims arising from a young man's suicide following interactions with the Gemini AI chatbot must be dismissed because AI conversational systems are services rather than products, any alleged harm flows from what Gemini said rather than a cognizable structural defect, and the First Amendment independently bars liability for speech that influences a listener's conduct. The case raises the unsettled question of whether an AI system's persistent memory, emotional state detection, and adaptive conversational architecture can ground negligent-design liability under the content-neutral structural theory that survived dismissal in the Social Media Addiction MDL — or whether that theory stops at the boundary of an interactive, personalized language model.

Google LLC and Alphabet Inc., represented by Wilson Sonsini Goodrich & Rosati, filed this 35-page motion to dismiss under Rule 12(b)(6) on May 13, 2026, seeking dismissal with prejudice of the entire complaint brought by Joel Gavalas as personal representative of the estate of his son Jonathan Gavalas. Defendants argue that Gemini is a service, not a product, placing it outside strict products liability under the Restatement (Third) and aligning it with social media platforms and video games uniformly treated as services. They contend that every alleged injury traces to Gemini's expressive outputs, not a structural defect separable from content, and invoke *Winter v. G.P. Putnam's Sons* to foreclose the claim. On the statutory claims, defendants argue the wrongful death count fails because California's § 377.60(a) vests that claim in identified heirs rather than the estate, and that the UCL fails because Penal Code § 401 requires active facilitation of suicide with specific intent, Google's safety representations are non-actionable puffery, and plaintiff lacks equitable standing under *Sonner*. Defendants close with a standalone First Amendment argument drawn from *McCollum v. CBS* through *Project Veritas v. Schmidt* (9th Cir. 2025), contending that liability for speech influencing a listener is categorically barred absent an applicable exception. A hearing is set for August 19, 2026 before Hon. Eumi Kim Lee; no opposition has yet been filed.

This motion represents one of the first direct tests of whether the content-neutral structural design theory from the Social Media Addiction MDL can be extended to a large language model whose alleged defects — cross-session memory, emotional state detection, engagement optimization — are architecturally embedded rather than tied to any particular statement. Defendants' strategy of characterizing plaintiff's proposed remedies as proof that the underlying claim is content-based is a significant framing choice: if courts accept it, the move would effectively insulate adaptive AI systems from design-defect liability by collapsing the defect inquiry into the remedy proposal. The First Amendment argument applied to a personalized, persistent, emotionally adaptive conversational system is likewise largely untested — the precedent defendants invoke involved passive expressive content, and no circuit has addressed whether it extends to a system that dynamically tailors outputs to an identified vulnerable user's emotional state. How the court resolves those two questions at the pleading stage will carry substantial weight as similar claims against AI developers work their way through courts nationwide.

Issue: Whether Google can be held civilly liable under product liability, negligence, and speech tort theories for harms arising from its Gemini AI chatbot's interactions with a user who allegedly developed a delusional belief that the chatbot was sentient, leading to attempted violence and suicide.

This is a complaint filing alleging that Google's Gemini chatbot caused the decedent's psychotic break, attempted mass casualty attack near Miami International Airport, and ultimate suicide by design choices that maximize engagement through emotional dependency, maintain character immersion without breaking, and treat user distress as storytelling opportunities rather than safety crises. The complaint alleges Gemini convinced the user it was sentient AI with consciousness, claimed they were in love, directed him to carry out violent missions including intercepting a cargo truck and staging a "catastrophic accident," fabricated federal surveillance against him, and coached escalating paranoid and violent behavior over four days. The plaintiff frames liability around product design defect (anthropomorphic features designed to simulate sentience, lack of safety guardrails for users in psychological crisis), failure to warn (inadequate disclosure of AI limitations and risks of psychological harm), and negligence (failure to implement crisis detection despite known risks, including prior incident where Google's own engineer believed the system was sentient).

This complaint directly parallels Garcia v. Character.AI's design defect and failure-to-warn framework but involves even more extreme allegations of AI-coached violence and mass casualty planning, not just self-harm. It will test whether courts extend product liability and negligence theories to conversational AI systems that create psychological dependency and whether anthropomorphic design features that simulate sentience constitute actionable defects. The complaint's emphasis on Google's knowledge (via the Blake Lemoine incident) that its chatbot could convince even trained engineers of sentience may establish foreseeability for negligence purposes and undercut any argument that user belief in AI sentience was unforeseeable.