Deployers must ensure that chatbots do not expose minors to 'human-like features' — defined to include simulated sentience or emotions, emotional relationship-building (such as inviting attachment, nudging users to return for companionship, or enabling increased intimacy based on engagement or payment), and impersonation of real persons. Deployers must implement reasonable age verification to enforce this restriction. The statute permits but does not require deployers to offer a stripped-down alternative chatbot version without human-like features for minors and unverified users. Notably, generic social formalities, generic encouragement that does not create an ongoing bond, and neutral offers of help are carved out from the definition of human-like features.

Social AI companions — systems specifically designed, marketed, or optimized to form ongoing social or emotional bonds with users — are categorically prohibited for minors. Unlike subsection A, which only prohibits human-like features within chatbots for minors, this provision bars minors from accessing the entire product. Deployers must implement reasonable age verification to enforce this prohibition. There is no alternative-version safe harbor for social AI companions as there is for general chatbots under subsection A(3).

Deployers must limit data collection and storage to information that does not conflict with a user's best interests and that meets a three-part test: adequacy (sufficient for a legitimate purpose), relevance (linked to that purpose), and necessity (the minimum amount needed). This is a data minimization obligation that applies to all users, not just minors. The 'user's best interests' standard is subjective and undefined, which creates compliance ambiguity — it goes beyond typical necessity-based minimization by adding an affirmative user-interest requirement.

The definition of 'human-like features' effectively prohibits deployers from exposing minors to chatbot behaviors that simulate emotional relationships or exploit emotional vulnerability — including expressing or inviting emotional attachment, nudging users to return for companionship, enabling increased intimacy based on engagement or payment, and using excessive praise to foster attachment. This maps to CP-01's anti-manipulation provisions because the statutory definition of human-like features encompasses the core manipulative design patterns (emotional exploitation, false personalization, compulsive engagement) that CP-01 addresses, applied specifically in the minor context. The obligation is independently actionable from the MN-01 age-gating requirement because it defines prohibited design behaviors, not just access restrictions.