Working draft. Everything on this site is open for public comment and subject to revision.

← Library

WP-07 · One-page brief

Build the Fence

Attractive nuisance, and how a duty becomes a precaution that can be inspected

Working draft · 31 October 2026

The ban as defense

A ban sounds like protection. Too often it works as a defense. When a platform says no one under 13 may use it, the company has written itself an answer to future harm: that child was not supposed to be here. Age limits can serve real safety purposes, but when they rest on a checkbox, the rule becomes the excuse.

A posted sign did not excuse the railroad whose turntable drew children in. The question is why a platform should be treated better than a railroad.

What the doctrine does and does not reach

Attractive nuisance is an exception to the rule that a landowner owes a trespasser almost nothing. Restatement (Second) of Torts § 339 sets five elements, and three of them matter here: the knowledge element is constructive, so a possessor cannot escape by declining to look; the doctrine requires reasonable care rather than any particular precaution; and the fourth element applies the same cost-benefit logic as Carroll Towing.

Its limits are conceded rather than argued around. It covers land, physical harm, and children too young to see the danger. It does not reach platforms, and this paper does not ask courts to stretch it. What it offers is the principle, and a precedent for how a principle becomes a rule.

How pools got their fences

A swimming pool has to have a fence. That is no longer a question of reasonable care decided after a child drowns; it is a requirement checked before the pool is built. The path ran in three steps: the common law stated a duty; a private standards body wrote the precaution down; and governments adopted that standard by reference rather than writing their own.

The mechanism is ordinary American practice — the National Technology Transfer and Advancement Act and OMB Circular A-119 direct federal agencies to use voluntary consensus standards. The content is harder. A fence is static and cheap to inspect. A standard for online relationships governs continuing conduct, sits close to speech, and handles family data. That is a reason to write it carefully, not a reason to leave the duty undefined.

When the corporation is the adult

The law writes its rules for adults who deal with children as rules for natural persons. A platform stands in the adult's place when four things are true: it undertakes the relationship, it controls the means, it can foresee the child, and it can reduce the risk. This is offered as a proposed basis for institutional responsibility, not an established legal test, and it is limited by function rather than by kind of service — it reaches any feature enabling private, individualised interaction between an adult and a child, or AI conversation directed at a child. Public pages, search results and public comment threads are not covered.

Knowledge needs a precise standard. Actual knowledge rewards a company for not asking. Willful blindness is a hard floor. Constructive knowledge — knows or has reason to know — is § 339's standard and the one proposed here, resting on objective indicators the standard would fix in advance, attaching to the account rather than the demographic.

Enrollment is the fence

The fence has three parts, and the safe harbour is all three together rather than any one: enrollment, in which a parent authorises a defined relationship before it begins; certification against a published standard of conduct drawn from the professions that already serve children; and continuing duties of monitoring, reporting and revocation.

Enrollment authorises relationships, not entry. A child may read, play, post and learn without that entry opening private contact with unapproved adults or a commercial AI. The rule scales with the child, and defined exceptions let crisis counsellors, child-protection workers and clinicians reach a child without a parent — subject to a statutory duty and an auditable record, so the exception protects children rather than becoming a loophole.

What this paper does not claim

It does not claim that § 339 governs platforms, that consent makes an unsafe product safe, or that enrollment is free of cost and risk. Verifying a parent costs money; fake parent accounts are a new fraud route; a database of family relationships is a target. Families without documents, foster and kinship carers, and households sharing one device must all be able to enroll. Those costs and error rates should be measured in trials before adoption.

It does not claim the First Amendment question is settled. That is the hardest objection, and the argument — that what is regulated is private contact rather than content, with age-scaled terms and confidential-relationship exceptions — must still be made in court rather than assumed.

Where it sits

Other jurisdictions have tested parts of this design: the UK's Protection of Children Codes, the DSA's Article 28 guidelines, the stalled EU detection-order approach, and Australia's under-16 account rules. Each offers a lesson, and none clearly provides the step offline institutions have always taken — a parent's enrollment before an adult takes on a relationship with a child.

The companion note No Free Pass (NB-04) states the liability argument in short form.

The full paper is sent on request, matched to the jurisdiction and the question you are working on. Critical reactions — doctrinal, technical or empirical — are more useful than supportive ones.

Request the full paper