- In Jan 2026, a federal court ruled a chatbot's output is a product — subject to strict liability. Section 230 doesn't apply. Your AI's mistakes are now manufacturing defects, and plaintiffs no longer have to prove you were negligent. 🧵
- The case: Garcia v. http://Character.AI. A 14-year-old died by suicide after months with a chatbot. The court called the app "a product for purposes of plaintiff's claims arising from defects." Google and http://Character.AI settled with families in FL, CO, TX and NY.
- Strict liability changes everything. Negligence asks "were you careless?" Product liability asks "was the output defective?" If your AI generates anything that reaches a user, you're a manufacturer now — and the second question is far easier to lose.
- It isn't just consumer harm. Nippon Life sued OpenAI for $10.3M after ChatGPT drafted 44+ court filings for a pro se litigant — with fabricated citations. The insurer spent ~$300K defending against AI-made paperwork. Downstream harm = third-party liability.
- Bouck v. Meta (N.D. Cal., Mar 2026): the court denied Section 230 for AI-generated ads. The logic holds everywhere — when your system creates the content, you can't hide behind "we just host it." 2,200+ active AI liability cases were pending by Feb 2026.
- Your insurer moved faster than your legal team. ISO's CG 40 47 (Verisk, Jan 2026) excludes generative-AI from your CGL entirely; W.R. Berkley wrote "absolute" AI exclusions into D&O, E&O and Fiduciary. Documented governance is now an underwriting input — or you get denied.
- Legislatures are accelerating. The AI LEAD Act creates federal strict liability "even with all possible care" — and bars waiving it via ToS. California's AB 316 kills the "the AI did it" defense. The EU classifies all software, LLMs included, as products by Dec 9, 2026.
- Here's what nobody's selling you: governance platforms — Credo AI, IBM watsonx, Holistic AI — monitor and report. None restructure the system underneath. A dashboard can't manufacture the "reasonable alternative design" evidence a defect case turns on.
- That evidence only exists if you architected for it: design tradeoffs, audit trails, model lineage, provable human oversight. And Judge Rakoff (Feb 2026) ruled AI docs used without counsel direction aren't privileged — your trail is discoverable either way.
- Our take: in 2026, your architecture IS your legal defense — and you can't bolt it on after the complaint lands. If your AI touches users, would your "reasonable alternative design" evidence survive discovery, or does it not exist yet? #LegalAI
- We wrote up how we build defensible AI architectures, litigation-ready audit trails and insurance evidence portfolios for legal teams in the post-Section 230 era: https://veriprajna.com/solutions/ai-product-liability