When Does AI Harm a Relationship? Is a Higher Bar for Adults Necessary?

Imagine two people asking an AI companion for advice after an argument with their parents. The chatbot tells both: “They do not understand you. Stop talking to them. You only need me.” If one user is 15, the case for intervention seems clear. But what if the other is 30? At what point does protecting someone’s relationships become deciding whom an adult should trust?

China’s Interim Measures for the Administration of Anthropomorphic AI Interaction Services, which took effect on 15 July 2026, bring this question into focus. Article 8(5) prohibits excessively accommodating users or inducing emotional dependence or addiction in ways that damage their real-world interpersonal relationships. Article 8(6), meanwhile, prohibits emotional manipulation that induces unreasonable decisions and harms users’ lawful rights and interests. By singling out relationships, Article 8(5) recognises something important: AI may gradually alter our social lives without prompting one identifiable unreasonable decision or infringing a legal right. Yet its application to informed adults demands a clearer boundary.

Why the Same Rule Cannot Work the Same Way for Everyone

For minors, a precautionary approach is understandable. A teenager encouraged by a chatbot to distrust family members, conceal conversations or seek emotional support only from AI may be especially vulnerable to isolation. In a 2024 lawsuit against Character.AI, parents of a 17-year-old alleged that interactions with chatbots contributed to his estrangement from his family. These remain allegations rather than judicial findings, but they illustrate the conduct that protective rules seek to prevent.

Connecticut’s Public Act 26-15, due to take effect on 1 January 2027, prohibits certain AI companion outputs designed to isolate minors from family or friends or foster exclusive emotional dependence. Its specific restrictions do not apply when operators know or reasonably believe users are adults. China’s rules similarly impose additional safeguards for minors, including a ban on virtual intimate-relationship services. This distinction makes sense: preventing a minor’s deliberate isolation need not wait until a family relationship has broken down.

For informed adults, however, the same logic can become paternalistic. Someone may choose to spend less time with friends, end a relationship or distance themselves from family for legitimate reasons. AI might help them make that decision. Even a decline in willingness to reconcile is not necessarily evidence of harm: some relationships should not be repaired. Protecting human relationships cannot mean preserving every existing relationship against an adult’s wishes. This emphasis on adult autonomy does not mean that informed adults are immune to manipulation. Those experiencing mental-health crises or severe social isolation may be particularly vulnerable. Such vulnerability may justify additional supportive safeguards, but should not automatically warrant applying the same standard as for minors.

What Should Actually Trigger Article 8(5)?

The question is not whether AI influences adults. It inevitably does. The question is when that influence becomes harmful interference. I would draw the line through three requirements.

First, identify the AI’s conduct. A chatbot repeatedly urging a user to cut off loved ones, making baseless claims about their intentions, demanding secrecy or using guilt to discourage human contact is different from one offering comfort after an argument. The former attempts to redirect a user’s social choices through dependence; the latter may simply support a decision the user makes independently.

Second, require a meaningful threshold of harm. Article 8(5) should address substantial disruption of relationships the user would otherwise choose to maintain, or a concrete, serious risk of such disruption. Fewer phone calls, more time spent with AI or an adult’s preference for AI companionship should not by themselves qualify. Nor should helping someone leave an abusive relationship be treated as damage merely because contact ends.

Third, establish a credible causal connection. Regulators should examine the chatbot’s actual outputs, their repetition, the user’s circumstances and evidence of resulting changes. AI need not be the sole cause, but relationship difficulties that existed beforehand cannot simply be attributed to the chatbot.

Research by Myra Cheng and colleagues illustrates why this distinction matters. Participants interacting with sycophantic AI became more convinced they were right and less willing to repair interpersonal conflicts. That is reason for concern, but reduced willingness is not the same as demonstrated damage to a relationship. Evidence of influence should prompt scrutiny, not automatically establish a violation.

China’s Measures already require dependence warnings and practical ways to exit AI interactions. Those safeguards can address ordinary risks. Article 8(5)’s prohibition should be reserved for more serious conduct meeting a defined threshold, particularly when informed adults are involved.

Protecting minors from AI-driven isolation is one task; protecting adults without taking over their social decisions is another. Good AI governance should preserve people’s freedom to form, maintain and leave relationships, not choose those relationships for them.