The difficult part of AB 1709 is not the politics. It is the fallback. A platform that has to stop personalizing for users under 16 must decide what those users see instead: a reverse-chronological river, a hand-built list of accounts, or a ranking model stripped of watch history and session length. Each option gives up engagement among an audience that supplies some of the heaviest time in the app.
That engineering bill is the quiet reason three of the largest platforms went to court rather than to their product teams. Meta, Google and TikTok sued California’s attorney general on Nov 5, asking a judge to keep AB 1709 from taking effect next year. The statute bars personalized feeds and autoplay for anyone under 16.
The three companies argue the law violates the First Amendment, and that it is preempted by Section 230 of the Communications Decency Act. The first claim is familiar. A feed is an arrangement of speech, and the state is dictating the arrangement. The second is the more interesting one, because Section 230 was written to shield platforms from liability for what users post, not to stop a state from regulating how a platform assembles what it shows.
Both arguments have travelled badly in recent years. Courts have grown more willing to treat recommendation systems as the platform’s own expression, which helps the companies in one direction and hurts them in another: if the ranking is their speech, the state has a stronger claim that they are speaking to children in a way it can regulate. The preemption theory asks a judge to read a liability shield as a design immunity. That is a larger step than the statute’s text obviously supports, and the platforms know it, which is why the complaint leans on the First Amendment first.
The timing of the filing is easier to explain than its legal theory. AB 1709 takes effect next year, and a challenge filed in November leaves room for a preliminary injunction before the compliance deadline. Suits filed in the spring would not.
Meta’s week was busier than the joint filing suggests. The company withdrew an update to its commercial AI terms that had been scheduled to take effect this month, and the UK Competition and Markets Authority opened a case in late October over its scraping of publisher content. Neither action is connected to AB 1709 on its face. Together they describe a company re-pricing the legal risk of several data practices at once, in several jurisdictions, at the moment those practices were about to become binding.
That pattern is the real subject. State legislatures have discovered that platform conduct can be regulated at the design layer, and the design layer is where the margins are. A rule about content moderation can be argued as a rule about speech. A rule about autoplay is a rule about a product feature, and it is much harder to describe as censorship without sounding like a company defending the right to keep a 14-year-old scrolling.
For the platforms, the cost of losing is not one state. California is the largest market, and a statute that survives a First Amendment challenge becomes a template that other legislatures can copy without paying for the litigation themselves. Compliance built for California would be portable, but the compliance itself is the expensive item, since it means maintaining a second ranking path and a second set of default settings for a slice of the user base that is hard to identify in the first place.
Identifying it is the awkward part. Enforcement depends on knowing which users are under 16, which means self-declaration, age estimation from behavior, or document checks. Self-declaration is unreliable and children have no incentive to be accurate. Estimation is an inference, and inferences are contestable in court. Document checks push a privacy problem onto teenagers and their parents, and would hand the platforms an argument that the state has effectively ordered identity verification for a lawful service.
The law’s defenders will make a narrower case than the headlines suggest. AB 1709 does not ban teenagers from the products. It bans two mechanics, and a chronological feed is still a functioning product. If a court accepts that framing, the First Amendment claim weakens, because the state is regulating a default rather than excluding anyone from speech. The counter is that a default is the product, and the platforms will spend the next year trying to establish that a feed without ranking is not a neutral arrangement but a demotion.
There is a second-order cost that neither side has an incentive to raise. Small developers cannot afford two ranking systems, and the largest platforms can. A rule written to protect teenagers from scale effects will, if it holds, push the marginal app toward the same few companies that can absorb the compliance expense. That outcome is not an argument against the law. It is the kind of thing that tends not to appear in a press release about child safety, and it will not appear in the briefs either.
What happens next is procedural. A judge decides whether to enjoin the statute before January, and the answer will turn on how the court characterizes a ranking model: a machine that sorts other people’s speech, or the platform’s own editorial judgment. The platforms have argued both positions in different cases, depending on which one helped. AB 1709 is the case where they have to pick one and live with it.
