Meta's $18B child safety deal: 'It allows Meta to define harm'

Meta's $18 billion settlement with 48 states is historic, but former engineering director Arturo Béjar says it has a fundamental flaw: "It allows Meta to define harm." The deal, announced Wednesday, imposes new safeguards on Instagram and Facebook for users under 18, including a two-hour daily limit, a midnight-to-6 a.m. block, and stronger age checks. Yet Béjar, the first witness in the federal trial the settlement cut short, argues that time limits alone don't address the addictive nature of content recommendations. "It's one thing to say, 'Yeah, you only get like two hours of alcohol or two hours of cigarettes a day,' but it's still as bad for you because of what's getting delivered," he said in an interview Wednesday.

The settlement, which also includes the District of Columbia and U.S. territories, requires Meta to pay out over a decade and to strengthen age estimation to keep kids under 13 off the platforms. It also mandates regular outside audits of age-checking technology and specific goals for false positive rates - minors identified as over 18. Meta agreed to use its own AI plus third-party tools, and if a user is kicked off for being under 13, the company will check the ages of their friends. Béjar, who testified last week that Meta took a "don't ask, don't tell" approach on underage users, praised the age assurance section: "It has measurement, has independent testing, has goals. Because I think Meta should be measured on effectiveness, not effort."

Many of the new safeguards are on by default, a shift from Meta's historical pattern of making safety features opt-in. The two-hour time limit can only be turned off with a parent's permission, and the night mode blocking use between midnight and 6 a.m. is also default. But other features, like turning off video autoplay or showing posts in chronological order instead of algorithm-driven feeds, remain opt-in. That distinction drew criticism from Josh Golin, executive director of online safety nonprofit Fairplay: "We are disappointed that the settlement does not turn off by default recommendation algorithms that connect kids to predators and send young people down dangerous rabbit holes."

The default-setting question surfaced at trial earlier this week when a lawyer for the states pressed Instagram head Adam Mosseri on the low adoption of the "take a break" feature. Very few teens used it when first introduced, but Meta made it default for teen accounts launched in 2024. Still, a teen could simply swipe away the notification. The settlement's effectiveness will take time to measure, said Yaël Eisenstat, director of policy and impact for the Cybersafety Research Center, which recently audited dozens of safety tools and found most fell short. "It is a little premature, unfortunately, to tell if exactly all the changes we want to see are going to happen," she said.

Eisenstat, who worked at Meta until 2019, called the settlement monumental in showing alternative approaches are possible. For years, Meta "decided to go against the advice of some of their own internal employees and continue to ignore the fact that these design features were unsafe in search of profit." She stressed that the measures at least prove technically feasible options exist - something whistleblowers have "been saying very loudly." But she added there's no "singular silver bullet," and progress will require technical solutions, market forces, and government regulation.

The settlement must still be approved by a judge, and Meta said the safeguards apply only in the United States, with most in place for 10 years. For executives across social media and consumer tech, the deal signals a regulatory shift: states are willing to extract binding commitments on design defaults, not just fines. The two-hour limit and night block are blunt instruments, but they set a precedent that platforms can be forced to change core UX for minors. The bigger unresolved question - how algorithms recommend content - remains untouched, leaving the door open for future litigation or legislation.

Béjar's warning that Meta gets to define harm is the crux. The settlement's metrics focus on age verification and time limits, not on the emotional impact of content delivery. That gap matters for any company facing similar scrutiny: if you don't define harm proactively, regulators will define it for you - and they may start with the most visible, measurable proxies. For now, Meta has bought a decade of regulatory breathing room, but the underlying critique from its own former engineer suggests the debate is far from over.