Panel 1: AI Companions: Design, Safety, and Governance

AI systems that act on behalf of users—whether navigating benefits, triaging legal questions, or providing emotional support—are rapidly moving from prototypes to deployment. These “loyal agents” and “AI companions” raise a shared set of questions: loyal to whom, safe for whom, and accountable how?

• What loyalty means in practice: Multi-step AI systems that take actions, call tools, and interface with users without constant supervision—from healthcare workflows and public-service delivery to emotionally salient, sometimes therapy-adjacent companionship products.
• From speech to service: When AI systems present as helpers or companions, consumer-protection, product-liability, child-safety, and mental-health frameworks become central. The Character.AI litigation illustrates how negligence and design-choice scrutiny may shape regulation.
• Design obligations and safety-by-design: Auditability, human-oversight patterns, age gating, dependency mitigation, disclosure, escalation to human support, and interaction with standards like the NIST AI Risk Management Framework.
• Enforcement and liability: State Attorneys General are emerging as key actors using UDAP, child-protection, and privacy law; litigation is functioning as de facto governance for companion AI.