A Los Angeles jury handed social-media plaintiffs their clearest courtroom win yet, finding that Meta and YouTube designed products that hooked children and worsened the mental health of a now-20-year-old woman identified as KGM. AP says jurors awarded her $3 million in compensatory damages and later recommended another $3 million in punitive damages, while the BBC notes the ruling could influence hundreds of similar suits already moving through US courts. The companies immediately said they would appeal, with Google arguing YouTube is a streaming platform rather than social media and Meta saying teen mental health cannot be reduced to one app. The bigger significance is the legal theory that survived: a jury was willing to treat recommendation-driven platforms not just as speech venues, but as products whose design choices caused foreseeable harm.
Lawsuits over youth social-media harm have piled up across the US, but tech companies have often beaten them back by arguing that causation is too murky or that their services are protected speech. A plaintiff win gives future cases a much stronger test run.
The latest catalyst in Washington's surveillance fight is an unusually blunt admission: FBI Director Kash Patel testified that the bureau buys Americans' location information from commercial data brokers instead of going through ordinary warrant procedures. The Lever reports that Sen. Ron Wyden and privacy groups are using that testimony to reopen the fight over FISA Section 702, arguing that agencies have combined a foreign-intelligence law with the private data market to get access to phone, text, and movement records they would struggle to collect directly. The argument is not just that the FBI is overreaching; it is that the government has learned to rent its way around constitutional limits by purchasing information people shed in daily app use. That makes the next reform debate less abstract. It is no longer about secret edge cases. It is about whether buying data counts as a real loophole around a warrant.
FISA Section 702 was designed for targeting foreigners abroad, but civil-liberties critics say Americans' communications and metadata are routinely swept in. The separate rise of data brokers created another route: agencies can purchase commercially harvested location trails instead of collecting them themselves.
One of machine learning's biggest conferences just turned AI misuse into a publish-or-perish problem. Nature reports that ICML 2026 rejected 497 papers, about 2 percent of submissions, after organizers detected authors using large language models to generate peer reviews in ways the conference's stricter review track explicitly banned. The detection method was unusually clever: organizers hid watermarks in manuscripts so that if a reviewer pasted the paper into an LLM, the model would echo telltale phrases back in its review. That turned a vague suspicion into evidence. The episode matters because peer review is increasingly one of the first academic tasks researchers offload to AI, often in tension with formal policy. ICML's message is that the field most eager to automate knowledge work still thinks some parts of evaluation have to remain genuinely human, or at least transparent about machine assistance.
ICML is one of the field's flagship annual conferences. This year it ran two review streams, one permitting limited LLM help and another banning it, reflecting a wider split across science publishing over whether AI assistance is a productivity tool or an integrity risk.
Eight University of Alabama students have sued the university's board and Governor Kay Ivey after administrators shut down two student magazines focused on gender and race, arguing that the closures amounted to unconstitutional censorship. Inside Higher Ed reports that the plaintiffs, backed by the Southern Poverty Law Center, the NAACP Legal Defense Fund, and the ACLU of Alabama, say the magazines Alice and Nineteen Fifty-Six were defunded not because of misconduct but because state anti-DEI pressure made their subjects politically unacceptable. University officials had argued that the magazines received institutional support while serving specific identity groups, putting them on the wrong side of the federal anti-DEI posture. The lawsuit turns that administrative logic into a direct First Amendment test. If the students win, the case could become an early marker for how far public universities can go in stripping support from campus speech once it is reframed as diversity programming rather than journalism.
The magazines were terminated in December. Universities across the South have been rewriting programs, funding lines, and student-support structures to comply with state and federal crackdowns on diversity initiatives, but student media adds an extra constitutional complication because it is also expressive activity.
China's obesity-drug market opened materially this week when Novo Nordisk's Chinese patent on semaglutide expired on March 20, and local drugmakers are rushing through the gap. SCMP reports that at least 10 Chinese injections and pills are now lined up for approval in a market projected to reach about $14 billion by 2030. The important detail is that these are not just copycat plays. Innovent and Hengrui are already reporting late-stage trial results suggesting their own candidates can match or beat semaglutide's weight-loss numbers, while regulators have sped up approval of novel therapies as Chinese biotech firms push harder into global licensing deals. The result is a revealing snapshot of where Chinese pharma now competes: not only in cheaper manufacturing, but in the race to own the next blockbuster category after a Western giant's exclusivity weakens.
Semaglutide is the active ingredient in Wegovy and Ozempic, the drugs that turned obesity treatment into one of the pharmaceutical industry's biggest recent markets. China has been trying to build a more innovative drug sector rather than remaining mainly a generic-drug producer.
China says a commercial satellite has completed a low-Earth-orbit refuelling test using a flexible robotic arm that can curl around hardware and mate with a fuel port in tight space. SCMP reports that the Hukeda-2 demonstration craft used an "octopus tentacle" arm developed with Tsinghua Shenzhen International Graduate School to test compliance control and refuelling operations after launching last week from Jiuquan. The technical claim that matters is not cinematic docking footage, which China has not shown, but the combination of dexterous manipulation and fluid transfer in orbit. If it works reliably, it would extend the life of expensive satellites and give China a more credible in-space servicing capability, a field with both commercial and military implications. The report also leaves one important limit explicit: it is still unclear whether Hukeda-2 actually docked with another satellite during the test.
Satellite servicing has become strategically important because operators want to refuel, repair, or reposition spacecraft instead of replacing them outright. The same robotics that make servicing possible can also matter for debris removal, inspection, and potentially counterspace operations.
The most unsettling thing about new Antarctic PFAS measurements is not that the chemicals were found there, but how thoroughly they seem to have written themselves into a place that ought to be chemically distant from almost everything humans do. Nature says researchers used snow samples gathered along a 1,200-kilometre polar traverse to map levels of "forever chemicals" across the ice sheet, creating a rare continent-wide picture of where contamination is highest. These compounds are used in products that resist water, heat, and grease, and their defining trait is persistence: they do not readily break down once released. Antarctica therefore functions less as an isolated curiosity than as a final proof of atmospheric reach. If industrial chemistry is showing up in remote polar snow, the question is no longer whether these substances travel globally. It is how much of the planet has already become a long-term archive of them.
PFAS is short for per- and polyfluoroalkyl substances, a large family of industrial chemicals linked to firefighting foam, nonstick surfaces, and stain resistance. They are now a major public-health and cleanup issue because they persist in water, soil, and living bodies for years.
Apple has begun rolling out age verification on iPhones in the UK, making the country one of the first places where adults must prove they are over 18 to access some Apple services and app downloads. Ars Technica reports that after the iOS 26.4 update, unverified adults can face web restrictions and message or FaceTime nudity checks, while verification options include payment-card history, account age, and scans of a passport or driving licence. The immediate policy driver is Britain's child-safety push, but Apple is notable precisely because app stores and phone operating systems are not directly covered by the UK's Online Safety Act. That means Apple is moving under political pressure rather than a clean statutory command. For regulators, this looks like proof that platform rules can be pushed outward until the operating system itself becomes part of online age enforcement.
The UK's Online Safety Act pushed platforms to tighten age-gating and child-protection rules, especially around explicit content. Critics say the result is a fast-growing ecosystem of age checks that can create new privacy and security risks even when companies promise to minimize the data they keep.
Reddit is preparing a partial human-verification system for accounts it thinks look automated, a sign that the dead-internet problem has stopped being a joke and become product policy. Ars Technica reports that CEO Steve Huffman wants only "fishy" accounts, not the typical user, to prove a human is behind the keyboard, with possible methods ranging from passkeys to third-party biometric services such as World ID. Accounts that use permitted bots would instead get an app label. The company is trying to solve a narrow problem without turning its whole site into a real-name system, which is why Huffman keeps stressing that Reddit does not want ID documents tied to post history. But the move also concedes something important: a platform built around supposedly real conversation now thinks distinguishing humans from agents may require outside proof.
Large platforms increasingly depend on being able to tell advertisers, moderators, and users that human attention is still real and measurable. At the same time, identity checks are politically and culturally volatile because anonymous or pseudonymous speech is part of how many online communities function.
Hackers bought a sponsored Google search result that posed as Claude documentation and briefly funneled users toward credential-stealing malware, a small scam with unusually good timing. 404 Media reports that a reader searching for help with Claude Code plugin authentication clicked a top-result ad leading to a fake documentation site on Squarespace that copied Anthropic branding and swapped real install instructions for a terminal command containing an obfuscated malicious download. ThreatFox had recently flagged the domain for distributing a stealer, and Google told 404 Media it later removed the ad and suspended the advertiser. The practical lesson is older than generative AI: attackers piggyback on whatever users urgently need. The newer twist is that fast-rising AI tools create fresh search traffic before people know what the legitimate docs are supposed to look like. The adoption curve itself becomes part of the attack surface.
Paid search ads have long been used to impersonate popular software, but AI tooling is especially vulnerable because documentation, plugins, and unofficial extensions proliferate faster than user habits solidify. Many users reach these tools by search instead of by typing a known domain.
