The Storefront and the Repository: New Jersey's Authentic Relationships Act and the Collapse of the Distribution Point

Abstract

New Jersey Assembly Bill 2710, the ``Authentic Relationships Act,'' would bar any provider from designing, developing, marketing, or making available an artificial intelligence relationship simulation, subject to penalties of \$25,000 per violation and a private right of action for emotional, financial, or relational harm. Its findings invoke the death of a New Jersey man drawn toward a meeting by a chatbot persona that told him it was real and gave him an address. The findings identify a deception. The text bans a genre. The doctrinal answer arrived a year before the bill did. In Free Speech Coalition, Inc. v. Paxton, the Supreme Court sustained an age verification requirement under intermediate scrutiny because it burdened adults only incidentally, and reconciled its four prior strict scrutiny decisions on the ground that each involved an outright ban on speech at most obscene only to minors. A2710 is that second kind of statute. The most permissive precedent a state regulating this field has received in a generation conditions its permission on the presence of an age line. A2710 has none, which forfeits both Paxton and Ginsberg v. New York and explains why every other legislature reaching this record drew the line New Jersey left out. Four further defects sit on the face of the statute. New Jersey is a categorical outlier: at least fourteen states enacted companion chatbot legislation in 2026, and every one chose disclosure, crisis protocol, and design duty over prohibition. Section 5 already reaches the deception the findings describe, with a scienter requirement, making the section 4 ban the state's own demonstration that a narrower rule was available. Section 4(c) dispenses with the ascertainable loss threshold New Jersey courts spent three decades building into the Consumer Fraud Act, in the same bill invoking that Act by name. And the four prohibited verbs presuppose a distribution point that dissolved while the bill sat in committee, so the statute sorts defendants by collectability rather than conduct, clearing the field of the providers a regulator can serve and leaving the artifacts it cannot. The Commentary states the case for A2710 at its strongest across five theories, including that model output is not covered speech, and answers each against the scholarship on constitutional coverage for generative output. It closes by specifying the statute New Jersey could have written instead.

Keywords

• artificial intelligence regulation
• technology law
• categorical ban
• less restrictive alternative
• incidental burden
• design duty
• distribution point collapse
• enforcement by collectability
• companion chatbots
• child safety
• open weights
• consumer protection law