Guidelines by field
AI guidelines for children and families what the field's own bodies ask
Companion chatbots became the fastest-moving area of AI law in the United States: from no state laws before 2025 to twelve by September 2026. International guidance and UK codes focus on design, defaults and privacy for anyone under 18.
Last reviewed 27 October 2026. Every claim links to its source.
At a glance
The child's best interests come first
In every design decision.
Always say it is not a person
All twelve US state laws require AI disclosure; California adds reminders every three hours for known minors.
Have a crisis protocol
Detect talk of suicide or self-harm and refer the child to real help.
High privacy by default
Protect children's data, with private settings as the starting point.
No designs that breed dependence
Avoid manipulative engagement, and check ages where content is harmful.
Who has spoken
- Binding law, regulation, court ruling or enforceable professional rule
- Voluntary guidance, principles or a code an organization chooses to follow
Voluntary Dec 2025
UNICEF
Guidance on AI and children, version 3.0
Requirements for AI that serves children's best interests, keeps them safe, protects their data and includes their voices.
Binding
UK Information Commissioner's Office
Children's code (Age appropriate design code)
Fifteen standards for online services likely to be used by children, which the regulator takes into account when enforcing data protection law.
Binding In force Jul 2025
Ofcom
Protection of Children Codes
Duties under the Online Safety Act, including age assurance where content is harmful.
Binding 2025 to 2027
Twelve US states
Companion-chatbot laws
In force in New York, California and Hawaii; nine more take effect in 2027. All require disclosure that the bot is AI and a crisis protocol. California lets people harmed sue for at least $1,000 per violation.
Binding Partly in effect since Mar 2026
State of California
Age-Appropriate Design Code
An appeals ruling let several provisions take effect, possibly including high-privacy defaults, which remain disputed, while others stay blocked.
Binding Sep 2025
US Federal Trade Commission
6(b) orders to seven companion-chatbot companies
Compulsory requests for information about effects on children and teens. A study, not a rule.
To take a break.
Binding and voluntary
Binding
State chatbot laws, California's design code provisions now in effect, the UK Online Safety Act codes and Children's code, and EU bans on AI that generates non-consensual intimate images or child sexual abuse material from December 2026.
Voluntary
UNICEF guidance. FTC 6(b) orders are compulsory information requests, not rules.
What changed in 2025 and 2026
- Twelve US states passed companion-chatbot laws by September 2026, and the December 2025 federal order on state AI laws carved out child safety.
- The FTC opened a study of companion chatbots and children in September 2025, and UNICEF issued the third version of its guidance in December 2025.
- Part of California's age-appropriate design law took effect after a March 2026 appeals ruling.
Where people still disagree
Age-checking is the sharpest disagreement: regulators see it as necessary where content is harmful, while critics say it asks too much of everyone's privacy.
Questions to ask yourself
- Does the child using this app understand, every time, that it is not a person?
- If my child told this chatbot they felt hopeless, what would it do?
- Have we, as a family, talked about when to ask a human instead?
How this lines up with my everyday guidelines
| My everyday guidelines | The field's nearest rule |
|---|---|
7. Model healthy use for the family Demonstrate balanced reliance: a tool, not a crutch. | Avoid designs that create emotional dependency (state chatbot laws; APA). |
7. Model healthy use for the family Show children how to think critically about AI responses. | Give children the skills to understand AI (UNICEF). |
7. Model healthy use for the family Normalize asking: does this seem right? Should we verify? | Always make clear that a chatbot is not a person (all twelve state laws). |