No Free Pass
Parental enrollment and liability for child-facing AI
Summary
Any company that forms a relationship with a child without the parent's enrollment should carry the liability for it, exactly as it would offline. The argument runs in six steps:
- A child cannot give permission or waive liability.
- A self-declared age is no proof anywhere else; the seller must check a credential.
- The only valid credential for a child comes from the parent: enrollment.
- Anyone who skips enrollment carries the liability.
- Certification is the safe harbor, as the safety mark and the good-faith ID check are offline.
- The medium earns no exemption. What is illegal offline should be illegal online.
Nothing here compels anyone. Liability attaches to the unenrolled relationship; certification and enrollment are the way out of it.
1. A child cannot give permission or waive liability
A child's click on a terms-of-service screen binds the company, not the child. Under the infancy doctrine, a minor may disaffirm a contract. The arbitration clause, the liability waiver and the consent language therefore do not hold against the child the way companies assume.
Data-protection law reaches the same result. Minors cannot consent to the processing of their data under COPPA in the United States or GDPR Article 8 in the EU; a parent must.
In many jurisdictions, even a parent cannot waive a child's injury claim in advance. Virginia goes further and treats pre-injury releases for personal injury as void generally. A company dealing directly with a child has no valid consent and no enforceable waiver behind it.
2. Self-declared age is no proof anywhere else
“I'm 18” opens the door online and nowhere else. A bar, a liquor store, a tobacco counter, a tattoo parlor and a lottery retailer cannot rely on a customer's word. The burden sits on the seller, not the buyer.
In most US states, the seller's defense to a sale to a minor is that it inspected a credential in good faith and the credential appeared valid. A seller that never checked has no defense.
Companies will argue that a child who lied about age cannot then disaffirm. Some states accept that estoppel argument and others do not. The split is itself the case for refusing self-declaration outright: a rule that turns on the child's honesty puts the burden on the child.
3. The parent's enrollment is the only valid credential
If the child cannot consent and the child's word is no proof, only the parent's act can authorize a relationship with a child. That is how every offline institution already works. A school, a camp, a doctor, a coach and a scout troop all receive the child from the parent, and anything beyond the agreed role needs a permission slip.
Enrollment is a better instrument than the ID check. Online age verification asks every adult to prove age to every service. Enrollment credentials the child once, at home, and asks adults to prove nothing. The service does not need to guess who is a child; it needs to know whether a parent has enrolled this child for this role.
For AI, enrollment also defines the role. The parent sets what the system is for, and questions outside that role go back to the parent.
4. Liability as the default, certification as the safe harbor
A company that forms a relationship with an unenrolled child carries the liability for it. A company that can show parental enrollment and certification for the role gets a presumption of due care.
The model is the product-safety mark. No statute made the safety laboratory's mark mandatory. Product liability did: manufacturers without it struggled to insure, retailers would not stock them, and juries punished them. Certification became the shield against a liability that already existed.
Enrollment and certification do different jobs, and a company needs both:
| Enrollment | Certification | |
|---|---|---|
| What it supplies | Permission for the relationship | Evidence of due care |
| Who gives it | The parent, only | An independent certifier |
| Without it | No permission at all | Permission, but no shield |
| Offline analogue | The permission slip | The safety mark; the good-faith ID check |
Certification stands where a waiver legally cannot. AI systems today act in any capacity and are certified in none; that gap is the plaintiff's theory of the case.
5. No free pass for the medium
The rules that govern a relationship with a child do not change because the relationship runs through a screen. A stranger cannot enroll a child in a camp, sell a child alcohol or tutor a child behind the parent's back. The same act online should carry the same consequence.
Online services operate free of rules that bind their offline competitors in the same activity. A liquor store must inspect a credential; a website takes a click. That asymmetry was never defended on its merits as a principle — it accumulated out of decisions taken in one country in the 1990s, and nothing in it binds the rest of the world.
The European Union already states the principle: what is illegal offline should be illegal online. That was the stated premise of the Digital Services Act, and it is the premise here.
Thomas Paine diagnosed this pattern in the opening of Common Sense (1776):
A long habit of not thinking a thing wrong, gives it a superficial appearance of being right, and raises at first a formidable outcry in defense of custom. But the tumult soon subsides. Time makes more converts than reason.
The asymmetry looks right only because of long habit. The industry's warnings that rules will stifle innovation are the outcry in defense of custom.
Paine also answered the claim to an exemption:
For as in absolute governments the King is law, so in free countries the law ought to be King; and there ought to be no other.
Exemption from the law is the old mark of royalty. A handful of technology companies now claim it: they ask to deal with children on terms no school, camp, doctor or shop is allowed, and call any rule that binds them a threat to innovation. That is a new royalty, one that answers to no law the rest of us obey. Paine's principle has no exception for code. The law is king, and there ought to be no other.
6. Enforcement design
The parent, not an agency, is the enforcer. Three design choices make that work:
- A private right of action for parents. COPPA has none; enforcement depends on the FTC and state attorneys general, and violations go largely unpursued. Illinois's biometric privacy law (BIPA) gives individuals the right to sue and has produced more real compliance than almost any US privacy statute.
- Statutory damages per violation. BIPA sets $1,000 per negligent and $5,000 per reckless or intentional violation, so plaintiffs need not prove dollar harm. Harm to a child is often real but hard to price; a fixed sum removes that obstacle. Illinois amended BIPA in 2024 to limit repeat recoveries, which shows the damages can be tuned.
- A trigger that is easy to prove. The violation is a relationship with an unenrolled minor. That is a fact in the company's own records, not a question of intent.
Insurance does the rest. Once the liability is real, insurers will price uncertified child-facing AI as high-risk, and certification spreads without anyone mandating it.
7. Jurisdictional notes
United States. Section 230 is the main barrier for platforms hosting third-party content. It is weaker for generative AI, where the output is arguably the company's own. In Garcia v. Character Technologies (M.D. Fla.), the court in May 2025 largely denied dismissal of product-liability and negligence claims over a 14-year-old's death and declined, at that stage, to treat chatbot output as protected speech. In Free Speech Coalition v. Paxton (June 27, 2025), the Supreme Court upheld Texas's age-verification law for sexually explicit sites under intermediate scrutiny. Commentators read that ruling narrowly, and challenges to broader parental-consent laws continue.
United Kingdom and European Union. Private suits by parents are a US-shaped tool. For these audiences, lead with regulator enforcement and collective redress, and with the DSA's own principle that what is illegal offline should be illegal online. GDPR Article 8 already requires parental consent for a child's data; enrollment is the instrument that supplies it.
8. Objections and answers
“Regulation stifles innovation.” This is the industry's standing answer to any rule. In May 2025, OpenAI's chief executive told the Senate Commerce Committee that requiring government approval to release powerful AI would be “disastrous” for the US lead, two years after naming a licensing agency as his top safety recommendation. In July 2025, Meta's global affairs chief refused to sign the EU's AI code of practice, calling it an overreach that would throttle frontier AI in Europe. In 2025, Amazon, Meta, Google and Microsoft backed a ten-year federal moratorium on state AI laws, arguing that a patchwork of state rules would stifle innovation.
The answer: an exemption from rules everyone else follows is not innovation; it is an unfair advantage. The school, the camp, the pediatrician and the liquor store all comply. A company that asks to deal with children on terms no offline actor gets is asking for the Wild West to continue. That has to end.
This proposal also requires no pre-release approval. It applies existing principles of consent and liability to a new medium. A company that wants to be predatory while everyone else has to be responsible will be held to account: sued by parents and, where statute provides, prosecuted.
“It will feed a litigation machine.” A real safe harbor answers this. Enrolled and certified means the claim fails early; only the unenrolled relationship is exposed.
“We can't know who is a child.” You don't need to. You need to know whether a parent enrolled this child for this role. Without enrollment, you carry the risk.
“Kids will lie about their age.” Offline law never lets the seller rely on that, and neither should online law.
Sources
- Hiett v. Lake Barcroft Community Ass'n, 244 Va. 191 (1992)
- Free Speech Coalition v. Paxton (US Supreme Court, 27 June 2025)
- Garcia v. Character Technologies (M.D. Fla.)
- Senate Commerce Committee hearing, 8 May 2025
- Paine quotations checked against the Project Gutenberg text of Common Sense
Scope and evidence
This note states the infancy doctrine, the BIPA damage figures, state ID-check defenses and the Digital Services Act framing from general knowledge; each requires a primary-source pin before it is relied upon. The industry statements in section 8 are drawn from contemporaneous reporting rather than from primary transcripts. The procedural history of Garcia v. Character Technologies is stated as of the May 2025 ruling and may have been overtaken since.
The full note is available as a PDF below, and may be circulated freely. Critical reactions — doctrinal, comparative or practical — are more useful than supportive ones.