Author: Nita Farahany Source: https://nitafarahany.substack.com/p/the-ninth-circuit-told-california Published: March 26, 2026
This is the closing installment of Week 9 of Nita Farahany's Advanced Topics in AI Law and Policy course (Class 9.3 of roughly 30), published as a Substack essay on March 26, 2026. It is a doctrinal walk-through of NetChoice v. Bonta II (9th Cir., March 12, 2026), the second time the same three-judge panel reviewed the California Age-Appropriate Design Code Act (CAADCA) within 18 months. Farahany uses the two decisions to draw a through-line for what survives constitutional challenge in child-safety legislation going forward.
First Amendment framing
Farahany argues that the First Amendment interest at stake is not platforms' asserted right to show children harmful content, but speakers' right not to have their speech filtered through the platform acting as a government-directed content screener under penalty of $7,500 per violation. The speakers she identifies as affected include teenagers, mental-health communities, news organizations, and LGBTQ+ youth.
NetChoice I (2024): the DPIA falls
In the first decision, the panel held that the Data Protection Impact Assessment (DPIA) requirement compelled non-commercial speech on matters of public controversy, triggering strict scrutiny. Farahany describes the requirement's "potentially harmful content" standard as broad enough to sweep in mental-health communities, school-shooting coverage, climate reporting, and teen grief posts. Because less restrictive alternatives were available, such as voluntary content filters and parental education, the DPIA requirement was permanently blocked.
NetChoice II (March 12, 2026): four questions, four outcomes
The second decision addressed four provisions and reached a different result on each.
- Coverage definition (California wins). The standard "routinely accessed by a significant number of children" is demographic rather than content-based. NetChoice failed to carry the facial-challenge burden under Moody. The question was returned to the district court for development of a full record.
- Age estimation (remanded). Because the provision includes an opt-out, platforms can apply child-level protections to everyone instead, which Farahany says substantially changes the constitutional analysis. The panel held Free Speech Coalition v. Paxton (2025) inapplicable on two grounds: it applies only to legally obscene content, and CAADCA triggers privacy defaults rather than gating access to content. The provision was sent back to the district court.
- Data-use restrictions and dark-patterns provision (blocked on vagueness). The phrases "materially detrimental to physical health, mental health, or well-being" and "best interests of children" were held unconstitutionally vague when applied at platform scale. Vagueness, not the First Amendment, was the doctrinal hook. Farahany notes that a redrafted provision with specific enumerated prohibited uses would carry a different constitutional profile.
- DPIA (still blocked). This provision remained blocked from NetChoice I.
The doctrinal through-line
Farahany distills a pattern across the two decisions. Provisions regulating settings, defaults, and enumerated data practices tend to survive, because they regulate commerce or conduct with at most incidental effects on speech. Provisions that require platforms to exercise content judgment, delegated by statute through vague standards backed by severe penalties, tend to fail.
She lists the CAADCA provisions currently operative under this pattern: default high-privacy settings for child users; the default profiling prohibition; precise-geolocation protections; age-appropriate disclosure; and the Children's Data Protection Working Group.
The drafting lesson she draws is that specificity is the constitutional path. A rule such as "use behavioral data to extend a minor's session duration" is, in her account, constitutionally easier to defend than "anything materially detrimental to well-being." She expects future child-safety law to be drafted with enumeration rather than open-ended standards.
Relationships
- part-of: Nita Farahany Advanced Topics course (Class 9.3 of ~30)
- related: NetChoice v. Bonta (CAADCA litigation), AI and the First Amendment
- previous: Inside My Advanced Topics Class 9.2: The Law That Kept Getting Blocked (Farahany, March 2026) next: Inside My Advanced Topics Class 10.1: The Government Can Take Your Blood. Can It Take Your Thoughts? (Farahany, March 2026)