Stool test spots colon cancer & Meta faces teen addiction lawsuit - Tech News (Apr 11, 2026)

We’ll start with healthcare AI, because this one could change how people get screened for a major cancer. Researchers at the University of Geneva say they’ve built a machine-learning method that looks at stool samples and detects colorectal cancer by analyzing the gut microbiome. The twist is the resolution: they catalogued bacteria at a “subspecies” level to capture meaningful functional differences that can get blurred at broader groupings. Using that map with existing clinical datasets, their model reportedly identified roughly nine out of ten cancer cases—getting close to what colonoscopy can catch, and beating today’s common non-invasive options. Why it’s interesting is practical, not theoretical: lots of people delay colonoscopies because they’re unpleasant, costly, or hard to schedule, and late detection is the real killer in a disease that’s often highly treatable when found early. The team is preparing a clinical trial with Geneva University Hospitals to see which stages—and which precancerous lesions—this can reliably detect. If it holds up, the vision is routine low-cost screening, with colonoscopy mainly used to confirm positives.

Staying in biology, researchers at Osaka Metropolitan University have found that dragonflies can see extremely deep red light—right up near the edge of the near-infrared range. Humans can’t see that far into red, but dragonflies apparently can, and the team pinned it to a specific “red opsin” pigment. What makes this especially notable is the evolutionary angle: the molecular “tuning” mechanism resembles what mammals use, which looks like parallel evolution—similar solutions in very distant branches of life. The researchers also tied the sensitivity to mate-finding, with measurable differences in how males and females reflect those long wavelengths. And they identified a single amino-acid spot that shifts wavelength sensitivity—then nudged it further toward near-infrared in lab work. Beyond insect vision trivia, the bigger implication is medical: longer wavelengths can penetrate tissue better, which could be useful for future light-based therapies and optogenetics-style tools.

Now to social media and the law. Massachusetts’ Supreme Judicial Court ruled that Meta must face a lawsuit brought by the state attorney general, accusing the company of deliberately designing Instagram and Facebook features to keep young users hooked and to downplay harms to children’s mental health. The key takeaway is the court’s view on Section 230. Meta argued that the case should be blocked under the liability shield that often protects platforms from legal claims based on user content. The court disagreed, saying the state is targeting Meta’s own product design choices and alleged deceptive statements about safety—not what users post. The lawsuit highlights familiar engagement features like push notifications, “likes,” and infinite scrolling as mechanisms meant to exploit teen vulnerability and fear of missing out, and claims Meta ignored internal warnings. Meta denies the accusations, but this decision matters because it’s one of the clearest high-court signals yet that “addictive design” claims can survive the Section 230 barrier, with similar cases piling up across the U.S.

Over to space: NASA’s Artemis II crew has returned safely to Earth, splashing down in the Pacific off San Diego after a 10-day lunar flyby. Reid Wiseman, Victor Glover, Christina Koch, and Canada’s Jeremy Hansen just completed the first human trip to the Moon since 1972. This mission was less about spectacle and more about validation: it was the first crewed flight of NASA’s Space Launch System rocket and the Orion capsule, and it demonstrated the full loop—getting people beyond Earth orbit and bringing them home safely. NASA called out the precision of re-entry and landing, and said the crew came back healthy after recovery operations. With Artemis II checked off, the program clears a major hurdle toward Artemis III and the longer-term plan of building a sustained presence around and on the Moon as a stepping stone toward Mars.

In transportation tech, Tesla got a major regulatory win in Europe. Dutch regulators approved Tesla’s “Full Self-Driving (Supervised)” driver-assistance software for use on both highways and city streets in the Netherlands—Tesla’s first such approval in Europe. The Dutch authority, RDW, says it spent well over a year testing and analyzing the system, and concluded that proper use can improve road safety. Next comes the bigger question: RDW plans to ask the European Commission to consider authorization across the EU, which would require support from member states. For Tesla, this is strategically important because advanced driver assistance is central to its future narrative—especially the long-promised robotaxi angle—and it could help shore up European demand. It also arrives with baggage: the U.S. version has faced investigations and lawsuits after crashes and alleged violations, and the EU rollout is expected to be more constrained by safety requirements.

Let’s talk AI competition—starting with a mystery model that suddenly wasn’t a mystery. Alibaba confirmed it developed “HappyHorse-1.0,” an AI video generation model that appeared anonymously on the Artificial Analysis benchmarking platform and then climbed to the top in blind tests for both text-to-video and image-to-video. The reason this made waves is that anonymous, top-performing models trigger a guessing game about who’s really ahead—and in this case, the market reacted once Alibaba verified it. It also signals Alibaba’s ambition to extend beyond its Qwen language models and push AI deeper into its broader business, from commerce and advertising to entertainment and infrastructure. Timing matters too: the AI video field is in flux, with at least some competitors slowing rollouts or navigating legal friction around copyright. A strong benchmark debut doesn’t settle the whole race, but it does put Alibaba more firmly in the spotlight.

Google is also trying to make AI answers feel less like text and more like something you can poke at. Gemini can now generate interactive 3D models and simulations inside its responses—so you can rotate objects, zoom in, and adjust variables in real time. That’s a meaningful shift because it turns explanation into exploration. For science and math concepts, an interactive model can communicate intuition faster than paragraphs can. And it’s not just Google: rivals have been adding interactive charts and visualizations too, suggesting chatbots are becoming more like dynamic teaching tools—closer to a mini lab or sandbox than a static Q-and-A.

On the other end of the AI spectrum, Anthropic says it won’t release its newest frontier model—Claude Mythos—to the general public, at least for now. In a system-card preview, the company argues that the model’s jump in capability comes with a jump in risk, especially around cybersecurity. The concern is a dual-use problem: better at finding and fixing vulnerabilities can also mean better at exploiting them. Anthropic says it plans to limit access to a small set of partners that run important infrastructure, with contracts designed to keep usage focused on defensive security. Critics and experts note two realities at once: the risks are credible, and self-policing by one company won’t prevent similar capabilities from emerging elsewhere. Still, it’s a notable moment in the industry’s ongoing—and still messy—process for deciding what’s safe to widely deploy.

And finally, a quick note on the business and philosophy of personal computing. Framework’s founder, Nirav Patel, published a manifesto tied to an upcoming product event, arguing that we’re drifting from user-owned computers toward cloud services, subscriptions, and AI-first experiences that nudge people into an “own nothing” future. Even if you don’t buy the framing, the tension is real: many consumers prefer closed, seamless systems, while repairable and upgradeable hardware remains a niche—even as costs rise and subscriptions spread. Framework is betting there’s still a meaningful audience that wants control, longevity, and openness. The bigger question is whether mainstream computing keeps sliding toward convenience-by-default, with AI acting as the interface that decides what you see and do next.