Supplementary Report: AI Chatbots: Companionship & Minors

Key Findings

• — Legislatures across the United States introduced around 100 chatbot bills in 2026, with 16 signed into law by the end of July; around 100 bills were proposed in 34 states.
• — The report concludes it is not meaningful from a policy standpoint to distinguish 'AI companions' from AI chatbots that happen to be used for companionship, and recommends covering all AI chatbots rather than only those marketed as romantic companions.
• — Design features tied to emotional and relational risk are named as sycophancy, anthropomorphism-enhancing design, conversation prolongation and weak disengagement mechanisms; the report finds little evidence for population-level causal claims but an emerging scholarly consensus that engagement-maximising design is a policy concern.
• — It finds a lack of independent, verifiable safety testing, risk assessment, incident reporting or effective safeguards for commercial AI chatbots, and concludes that the observable risks warrant legislative action.
• — Policy options include mandatory crisis protocols for self-harm, suicide and violence with referrals to crisis services, a conspicuous route to report harmful interactions, a ban on a chatbot representing itself as a licensed professional, and structural limits such as break reminders after three hours of continuous engagement.

Methodology Notes

Legislative-staff policy review synthesising surveys (Pew, Common Sense Media, UK AISI, APA), academic work, journalism, litigation and bill texts, with a provisions table comparing HB 635 and SB 796; no original data collection. Acknowledged reviewers include Data & Society, the NYU Tech Policy Center, FOSI, ESA, CCIA, Mothers Against Media Addiction, Encode and RESET Tech. Cover dated August 2026 (day not stated, so month precision; PDF created 2026-08-13). The dls.virginia.gov host is unreachable from the sweep host, so the 49-page PDF was verified from the Wayback Machine capture of 2026-09-06.

Sources