The Feed Is the Platform Safety Fight

The safest legal move in the platform wars is also the least honest one: pretend the feed is somehow separate from the product.

I understand why regulators and courts are tempted by that move. If the government tells Instagram, YouTube, TikTok or X which lawful posts to suppress, boost or carry, it walks straight into the First Amendment. In the United States, that problem sits on top of Section 230¹, the 1996 law that generally stops courts from treating online services as the publisher or speaker of other people’s posts. The old content-moderation fight was about whether platforms take down too much or too little speech. The new fight is harder: whether platforms can be held responsible for the way their products steer children, vulnerable users and political communities toward harm.

My answer is yes, but only if we stop pretending platform safety can be regulated while leaving the feed untouched. The feed is not a bulletin board with some optional software sprinkled on top. On TikTok’s For You Page, YouTube’s homepage, Instagram Reels, Snap’s discovery surfaces and X’s timeline, an algorithmic recommendation system selects, ranks and repeats content based on predicted interest, prior behavior and platform goals. That system is the product.

That does not mean the state should become a feed editor. It does mean any serious duty of care, meaning a legal obligation to take reasonable steps against foreseeable risks, has to include the recommender system. The workable path is feed accountability without viewpoint control: audits, risk testing, minor-protective defaults, limits on engagement-based profiling of children, and liability for reckless amplification after clear notice of severe risk. Laws that stop at privacy settings and parental dashboards will help around the edges. They will not change the machine.

The legal system is already circling this distinction. In Moody v. NetChoice², the Supreme Court treated Facebook’s News Feed and YouTube’s homepage as curated expressive products, warning that state laws forcing platforms to carry or avoid demoting certain viewpoints burden editorial judgment. That is the strongest argument against crude feed mandates. A law that says “do not demote this political viewpoint” is not child safety law. It is speech control.

But Moody did not answer the full product-safety question. Product liability, the legal framework used when a product is defectively designed or sold without adequate warnings, does not usually ask whether a product expresses an idea. It asks whether the maker created an unreasonable risk. That is why the social-media addiction cases matter. In the federal multidistrict litigation against Meta, Google, TikTok and Snap, a California federal court allowed some design-focused claims to proceed while treating claims tied to third-party content publication more skeptically; surviving theories included issues such as parental controls, barriers to account deletion, failure to warn and some platform-created notifications, according to the court’s order summarized in FindLaw’s copy of the ruling³.

The recommendation cases are messier. In Anderson v. TikTok⁴, the Third Circuit let claims proceed over TikTok’s alleged recommendation of the “Blackout Challenge” to a child, reasoning that the For You Page recommendation could be treated as TikTok’s own expressive activity rather than merely hosting someone else’s video. In M.P. v. Meta⁵, the Fourth Circuit went the other way, holding that Facebook’s arranging and sorting of third-party content remained a publishing function protected by Section 230. That split is the map. Courts are trying to decide when ranking is speech publishing and when it is defective product conduct.

The recent verdicts show why this will not stay theoretical. On March 25, 2026, a Los Angeles jury found Meta and YouTube negligent in a bellwether social-media addiction trial and awarded $6 million to a young woman who alleged that Instagram and YouTube harmed her mental health through addictive design, according to The Guardian⁶ and TechCrunch⁷. On May 15, 2026, Reuters reported that YouTube, Snap and TikTok reached settlements in school-district litigation seeking to recover costs tied to alleged youth mental-health harms from social-media addiction, as carried by Investing.com⁸. These are not content-moderation suits in the old sense. They are attempts to make platforms answer for design.

The evidence is not simple, but it points to design. The U.S. Surgeon General’s advisory says children and adolescents who spend more than three hours a day on social media face double the risk of mental-health problems such as symptoms of depression and anxiety, while also acknowledging that social media can benefit some young people through connection and support, according to HHS⁹. The National Academies’ 2024 report put it more carefully: the science suggests some social-media features can harm some young people’s mental health, but the evidence varies by user, context and design, according to the National Academies¹⁰.

That nuance matters. A court should not infer that every sad teenager was injured by Instagram. But a regulator should care when design repeatedly narrows a child’s world. The Center for Countering Digital Hate’s TikTok audit, which should be read as an advocacy-group test rather than population-level proof, found that teen test accounts were recommended suicide content within minutes and eating-disorder content soon after, with accounts marked as vulnerable receiving far more self-harm and suicide recommendations, according to CCDH¹¹. The exact numbers can be debated. The mechanism cannot: if a recommender learns distress and feeds it back, the harm is not just that bad content exists. The harm is repetition, sequencing and personalization.

Platforms have already conceded this in practice. TikTok announced a default 60-minute screen-time setting for users under 18 in 2023, while still allowing teens to continue after entering a passcode, according to TikTok’s newsroom¹². Meta’s Instagram Teen Accounts put teens into private accounts by default, restrict messages, use sleep mode from 10 p.m. to 7 a.m. and require parental permission for younger teens to loosen some settings, according to Meta¹³. YouTube said it would limit repeated recommendations to teens of videos involving body comparison, idealized weights or fitness levels and social aggression, according to TechCrunch¹⁴. These changes are important because they are not merely about reporting buttons. They alter what gets repeated.

Europe has stopped pretending the feed is outside regulation. The European Union’s Digital Services Act, or DSA, requires very large online platforms to assess systemic risks arising from the design or functioning of their services, including algorithmic systems and recommender systems, under Article 34¹⁵. In February 2026, the European Commission preliminarily found TikTok in breach of the DSA over “addictive design,” identifying infinite scroll, autoplay, push notifications and a highly personalized recommender system, and saying TikTok may need to adapt its recommender system, according to the Commission¹⁶.

The United Kingdom’s Online Safety Act also uses a systems-and-processes model, with Ofcom’s children’s safety materials identifying recommender systems as risk factors and requiring services to assess and mitigate risks to children, according to Ofcom¹⁷. Australia chose a blunter route: as of December 10, 2025, platforms including Facebook, Instagram, Reddit, Snapchat, TikTok, Twitch, X and YouTube must take reasonable steps to prevent Australians under 16 from having accounts, according to the eSafety Commissioner¹⁸. That may reduce some exposure, but it also pushes the fight into age assurance, evasion and privacy.

The United States is experimenting in fragments. New York’s SAFE for Kids Act prohibits addictive feeds for minors and includes overnight notification rules, according to the state-law text collected by New York Public Law¹⁹. California’s SB 976 targets addictive feeds, notifications and other features for minors, and the California attorney general announced in September 2025 that the state had secured an early win against NetChoice’s challenge, according to the California Department of Justice²⁰. Those laws will still face hard constitutional tests. But they are aimed at the right layer.

Privacy is the trapdoor. End-to-end encryption means only the communicating users can read message contents, not the platform. That protects journalists, dissidents, children and ordinary people from surveillance. It also makes detection of abuse harder. The National Center for Missing & Exploited Children reported 29.2 million suspected child-exploitation incidents for 2024, down from a comparable 36.2 million reports in 2023, and identified broader use of end-to-end encryption as one factor in the decline, according to NCMEC²¹. A safety law that responds by forcing universal scanning or encryption backdoors would create its own danger. The better answer is mandatory abuse-reporting workflows, fast escalation, evidence preservation after user reports, transparency about detection limits and risk audits that do not require the government to read everyone’s messages.

The strongest counterargument is that once regulators touch recommender systems, platforms will over-remove lawful speech to avoid liability. I take that seriously. A vague duty to prevent “harmful content” would pressure companies to suppress controversial speech, especially around politics, sexuality, war, public health and religion. The cure is specificity. Do not ask juries to decide whether a platform should have shown a lawful political post to fewer people. Ask whether the company tested foreseeable minor-safety risks, whether it kept engagement maximization as the default for children after finding severe risks, whether it gave researchers meaningful access, whether it reduced repeated exposure to self-harm spirals, and whether it acted after documented notice of a dangerous recommendation pattern.

So the answer to the headline question is no. Platform safety cannot be regulated well without regulating the feed itself. But the right feed regulation is not government ranking. It is product governance: prove the system was tested, disclose enough for outsiders to verify the test, set safer defaults for minors, and face liability when a platform knowingly keeps a dangerous amplification loop because it is profitable.

The indicator to watch is whether the European Commission’s TikTok case, California’s addictive-feed law and the next U.S. social-media addiction trials produce remedies that require measurable recommender changes rather than more parental-control dashboards. If by the end of 2027 the major platforms have not adopted independently audited minor-safety tests for recommendation sequences, I would expect the next wave of law to become blunter, with more age bans, more litigation and worse privacy tradeoffs.