Meta has agreed to put hard limits on how teenagers use Instagram and Facebook in the US, including a two-hour daily cap that teenagers can’t disable themselves and an overnight block on much of both apps. Most push notifications will be silenced during school hours, like counts will be hidden by default, teenagers will be able to choose a feed that isn’t personalised by Meta’s recommendation systems, and parents will get stronger controls over how their children use the platforms. The changes are contained in a proposed agreement between Meta and 52 US attorneys general and will apply to under-18s in participating US states and territories if the settlement receives judicial approval.
Meta’s version of how it arrived here is unsurprisingly very generous to the company which released an announcement describing the agreement as an extension of its existing work to protect teenagers and repeatedly calls it a new “industry standard”, while urging TikTok and YouTube to follow its lead. The states reached this agreement with Meta after litigation had already progressed to trial, following allegations that the company made design choices that drove excessive use among children and teenagers, collected data from under-13s unlawfully, and misled families and the public about the safety of Facebook and Instagram. Meta hasn’t admitted wrongdoing, but the protections weren’t conceived in a vacuum or introduced without considerable legal pressure behind them. California Attorney General Rob Bonta’s account of the lawsuit and settlement says the trial began on 18 August, little more than a week before Meta published its call for the rest of the industry to join it.
The conditions attached to the deal make Meta’s appeal to its competitors far more interesting. Meta says the agreement involves up to $17.1 billion paid over a decade, of which about $12.7 billion is allocated regardless of what happens elsewhere in the industry. The remaining roughly $5.3 billion will only be released if TikTok and YouTube implement specified one-hour limits, Night Mode and age-assurance measures, and both companies make matching payments. New York’s attorney general calculates the arrangement differently, describing a minimum payment of $12.1 billion and a maximum of $17.1 billion if other major social-media companies reach similar settlements, so the safest description is that several billion dollars of Meta’s eventual liability is conditional on competitors being brought into comparable agreements.
Some of the protections work the same way. Meta’s initial daily limit is two hours cumulatively across Facebook and Instagram, while Night Mode blocks access to feeds, Stories, Explore and Reels between midnight and 6am. Those particular commitments initially last for five years. If its industry peers sign on, Meta says the commitment stretches to ten years, the daily restriction becomes one hour per app and Night Mode expands to run from 10pm until 7am. Direct messages remain exempt from the time, school and overnight restrictions, allowing teenagers to continue communicating even when other parts of the services are unavailable.
There is a defensible policy argument behind wanting TikTok and YouTube covered by similar rules. Teenagers move between services, so a hard limit on Instagram can shift time to TikTok instead of reducing overall social-media use, while a company facing restrictions its rivals don’t face could also be commercially disadvantaged. Meta’s call for equivalent standards across large platforms therefore isn’t unreasonable on its own. The settlement goes further, though, because the level of protection Meta gives its own users changes according to whether its competitors accept similar obligations. If Meta and the states believe 10pm is a safer point to restrict Instagram than midnight, or that one hour is preferable to two, the safety effect on an Instagram user isn’t altered by YouTube’s legal position. Meta’s competitive position is.
There’s no public evidence showing that Meta demanded this particular settlement structure, so we can’t know how the conditional terms were negotiated. The terms themselves are revealing enough though. As mentioned earlier, Meta is committing to some restrictions for five years, but will extend parts of that commitment to ten years if TikTok and YouTube accept comparable obligations. Some protections also become stricter under the same condition. That means Meta’s willingness to provide the stronger version of its own teen-safety regime is directly linked to whether its competitors have to do the same. If those restrictions are better for teenagers, it’s difficult to see why their value should diminish after five years simply because TikTok or YouTube hasn’t signed an equivalent agreement.
Instagram’s safety controls have a history
Meta has spent years introducing safeguards for younger users, and some of those protections have become substantially stronger. Teen accounts are private by default, younger users are placed into more restrictive messaging and content settings, sleep mode silences notifications overnight, and teenagers under 16 need parental approval to weaken several controls. Meta launched Instagram Teen Accounts in September 2024 and completed their global rollout in 2025, later saying 97% of 13-to-15-year-olds kept the built-in restrictions they had been given.
The question isn’t whether Meta has ever built safety features. It is whether those protections were strong enough, arrived early enough and changed the parts of Instagram most capable of driving prolonged or harmful use. Meta’s earlier safeguards included Take a Break, Nudges, Hidden Words and parental supervision, alongside restrictions on unwanted contact and controls intended to reduce exposure to sensitive content. Take a Break is particularly revealing because Instagram launched it in 2021 as an optional feature intended to prompt people to leave the app after scrolling for a period they chose themselves.
Evidence presented during the litigation showed how little reach that approach initially had. Instagram head, Adam Mosseri, was questioned about an internal document showing that only around 1.8% of teenagers enabled Take a Break early in its life, despite Instagram publicly highlighting that more than 90% of teenagers who had enabled the feature continued using it. Mosseri argued that focusing on one feature ignored the broader safety work Instagram was doing, but he also acknowledged that Take a Break remained little used before Meta eventually made it part of the default experience for teen accounts.
Former Meta data scientist George Volichenko’s evidence went further. Volichenko, who worked on Instagram’s mental-wellbeing team, testified that the team wanted Take a Break switched on automatically for younger teenagers because an opt-out feature would reach far more people than one that had to be deliberately enabled. The proposal wasn’t approved, and Volichenko told the court that the effect on Meta’s core usage metrics was considered undesirable. Meta disputes the broader allegation that it deliberately held back safety work, while Mosseri has argued that Instagram evaluates safety across many different interventions rather than a single control. Even with those qualifications, the episode leaves an obvious weakness in the argument that providing a tool is equivalent to solving the problem the tool addresses.
Independent testing has raised similar doubts about some of Meta’s later protections. In 2025, child-safety groups working with researchers at Northeastern University tested 47 Instagram safety features and judged only eight completely effective. They reported weaknesses in restrictions intended to stop teenagers finding self-harm material, failures in some anti-bullying filters and problems with systems meant to interrupt repeated consumption of harmful content. Reuters reproduced some of the weaknesses independently and reviewed Meta documents showing that employees had identified problems with some detection systems, while also finding that controls such as Quiet Mode and parental approval for settings changes worked as intended. Meta called the report misleading and said teenagers placed into its protections encountered less sensitive material, received less unwanted contact and spent less time on Instagram at night. Reuters’ investigation into the performance of Instagram’s teen-safety features gives enough evidence to reject both extremes: Meta’s safeguards aren’t worthless, but their existence hasn’t guaranteed that they work as advertised.
The research is more complicated than Meta’s critics sometimes suggest
There’s no evidence around social media and teenage mental health to support the claim that Instagram inevitably harms everyone who uses it. That was true when Meta’s internal research became public in 2021 and remains true after several more years of study. One of the most widely repeated findings from the company’s own research was that 32% of teenage girls who already felt bad about their bodies said Instagram made those feelings worse. Meta objected to broader reporting around the documents and pointed out that across most of the wellbeing issues it examined, struggling teenagers were more likely to report that Instagram made things better rather than worse. Body image was the conspicuous exception.
Later independent work has produced a similarly mixed picture rather than a clean verdict. A 2025 longitudinal study in JAMA Network Open followed 11,876 children and adolescents and found that increases in an individual’s social-media use were associated with greater depressive symptoms a year later, while depressive symptoms weren’t associated with subsequently spending more time on social media in the same analysis. The authors didn’t claim that time on social media alone explained depression, and previous longitudinal findings haven’t all pointed in the same direction, but the study strengthens the case for treating prolonged use as a meaningful health question rather than a neutral measure of engagement.
The US Surgeon General’s advisory on social media and youth mental health cites research showing that young people who spend more than three hours a day on social media face about double the risk of mental-health problems including symptoms of anxiety and depression, while 46% of surveyed 13-to-17-year-olds said social media made them feel worse about their body image. Those are associations rather than proof that social media caused an individual teenager’s mental-health problems, but they sit uncomfortably beside a business model that benefits when users remain engaged for longer.
The National Academies’ 2024 review of social media and adolescent health was more cautious about population-wide claims. Its committee concluded that the evidence didn’t support attributing broad changes in adolescent health to social media alone and also recognised benefits such as community and connection. It nevertheless identified particular platform characteristics that can harm some young people, including algorithmically amplified harmful material and the displacement of healthier activities. The science therefore gives regulators a stronger basis for examining product design than it does for declaring social media uniformly harmful.
Facebook chose what its algorithms should reward
Algorithmic feeds belong in this discussion because they are frequently described as though the software itself decided what Facebook should value. Facebook launched News Feed in 2006 and progressively developed ranking systems to decide which posts users should see most prominently. By 2021, Meta was describing a highly developed machine-learning system that predicts which content someone is likely to find relevant or engage with and then scores competing posts to determine their position in the feed. Meta’s technical explanation of News Feed ranking makes clear that machine learning performs the prediction, while the company still chooses the signals and outcomes used to calculate what receives priority.
Some of those choices produced consequences Facebook later had to manage. Internal company documents reported by the Washington Post showed that from 2017 Facebook initially treated reaction emojis, including the angry reaction, as five times more valuable to its ranking system than an ordinary like. The logic was that reactions represented stronger engagement, but Facebook’s own researchers later found angry reactions appearing disproportionately on problematic material including misinformation and toxic content. Replies could receive even greater weighting as Facebook tried to encourage what it called meaningful social interactions, helping posts that made people angry or offended attract the activity its system had been designed to reward. The Washington Post’s reporting on Facebook’s reaction weighting showed that when the company eventually reduced the weighting of anger, its own data indicated that users saw less disturbing material and misinformation.
Walt Mossberg had been asking uncomfortable questions about Facebook long before its recommendation systems became a subject of regulatory scrutiny. Mossberg, the hugely influential former Wall Street Journal technology columnist who later co-founded AllThingsD and Recode with Kara Swisher, pressed Mark Zuckerberg in 2010 over Facebook’s approach to privacy and its decision to switch on Instant Personalization rather than asking users to opt in. Zuckerberg defended that approach partly by pointing to News Feed, another Facebook feature that users had initially resisted before eventually accepting.
The more direct warning about Facebook’s algorithms came several years later from Roger McNamee, an early Facebook investor and former adviser to Zuckerberg. In 2016, McNamee became convinced that there was a systemic problem with Facebook’s business model and algorithms that could allow bad actors to manipulate users. He first took those concerns to Swisher and Mossberg at Recode. Mossberg encouraged him to write an opinion piece, and McNamee subsequently sent a draft directly to Zuckerberg and Sheryl Sandberg warning that Facebook’s algorithms and business model could be used to harm people.
That history makes it harder to treat the problems regulators are confronting now as consequences nobody could reasonably have anticipated. The warnings changed over time, from Mossberg questioning Facebook’s habit of making consequential choices on users’ behalf to a former Zuckerberg adviser explicitly identifying the incentives built into Facebook’s algorithms.
Seen against Meta’s new settlement, those earlier warnings take on a different significance because many of the protections now being imposed are built around defaults. Teenagers don’t simply receive another reminder that they can use Instagram less; a time limit is imposed unless a parent changes it. Like counts disappear automatically, Night Mode blocks large parts of the service during specified hours, and parents can require their children to use a feed that isn’t personalised by Meta’s recommendation systems. The settlement accepts, through its design, that defaults exert enough influence over behaviour that leaving every protective decision to the person using the product is inadequate.
Meta has been in trouble over young users repeatedly
The current settlement is part of a much longer run of regulatory action and litigation involving children, teenagers and the safety of Meta’s products. Counting distinct major proceedings rather than every hearing, ruling or appeal within the same case, there have been at least six substantial legal or regulatory tracks involving young users since 2022. They concern different laws and different forms of alleged harm, so they shouldn’t be presented as six findings that Instagram causes the same problem, but the frequency makes it difficult to argue that this week’s intervention responds to a newly discovered concern.
In September 2022, Ireland’s Data Protection Commission fined Meta €405 million following an Instagram inquiry into children’s data. The investigation covered the public disclosure of children’s email addresses or phone numbers through business accounts and Instagram’s use of a public-by-default setting for children’s personal accounts. Less than a year later, the US Federal Trade Commission accused Meta of failing to comply fully with an earlier privacy order and misleading parents about whom children could communicate with through Messenger Kids, proposing restrictions that would have prevented Meta from monetising data collected from under-18s.
The multistate case that produced the new settlement was filed in 2023, alleging that Meta used features and recommendation systems that encouraged young people to remain on its platforms while presenting those services as safer than the states believed they were. New Mexico pursued its own case and this month secured a final judgment that brings Meta’s financial exposure there to $942 million, combining a $375 million jury penalty with a further $567 million order and court-supervised reforms. Meta has said it will appeal.
Meta also lost a separate Los Angeles case in March alongside Google’s parent company Alphabet. A jury found the companies negligent in the design of Instagram and YouTube and awarded $6 million to a woman who said she became addicted to the services while she was a child, allocating $4.2 million of the damages to Meta. Both companies said they planned to appeal. The case is significant because the argument focused on product design, including features such as infinite scroll, rather than relying solely on the harmful content posted by other users.
European regulators have meanwhile been examining whether Meta’s own protections are sufficient under the Digital Services Act. In April 2026, the European Commission preliminarily found that Facebook and Instagram weren’t doing enough to prevent under-13s from accessing the platforms, pointing to the ease with which a child could enter a false date of birth. In July, the Commission separately reached a preliminary finding that the addictive design of Facebook and Instagram breached the DSA, focusing on infinite scroll, autoplay, push notifications and highly personalised recommendation systems. The Commission said Meta’s existing mitigation measures hadn’t effectively dealt with the risks its investigation identified. These remain preliminary findings rather than final judgments, but they are difficult to square with a narrative in which Meta’s existing protections had already settled the issue.
The six tracks aren’t identical, and some remain contested or under appeal. What they establish is a repeated pattern of regulators, governments and plaintiffs challenging the adequacy of Meta’s treatment of young users while the company continued adding safeguards of its own. Seen across that record, Teen Accounts look less like evidence that outside intervention had become unnecessary and more like one stage in a process in which Meta’s definition of adequate protection has kept changing as pressure around the platforms has increased.
South African teenagers won’t get the new US limits
Teen Accounts already apply in South Africa because Instagram completed their global rollout in 2025, but the hard limits contained in the new settlement won’t automatically follow them here. Meta says the agreement applies to under-18s in participating US states and territories, which means a teenager in South Africa can receive the existing Teen Account protections without receiving the settlement’s two-hour hard cap or the new restrictions attached to US users.
That geographical difference deserves attention because the debate over young people’s access to technology is already moving beyond the idea that parents should simply be given another settings page. The same tension appeared in our recent examination of governments restricting social media for teenagers while allowing them access to increasingly powerful AI tools. Australia has already adopted a much more restrictive approach to social-media access for under-16s, while the US settlement takes a different route by allowing access but placing limits inside the products themselves.
TikTok and YouTube should face the same scrutiny over the systems they use to hold young people’s attention, and Meta is entitled to argue that regulating one platform while leaving equivalent practices untouched elsewhere can produce poor policy. The conditional terms of its settlement still leave Meta with a problem of its own making, because the company has accepted that stricter time limits and longer overnight restrictions are appropriate enough to include in a child-safety agreement while making their strongest form dependent on competitors joining it.
After years of optional controls, increasingly restrictive teen accounts, and recurring clashes with regulators and courts, Meta is finally accepting limits that reach into the core mechanics of its products. Whether these restrictions actually work remains to be proven, especially given its track record of safeguards that looked better on paper than in practice. But Meta’s core contradiction is harder to defend: demanding that TikTok and YouTube meet the same standard doesn’t explain why safety on Instagram should ever be conditional on what its competitors are willing to spend.