AI News - Apr 10, 2026

Welcome to AI News in 5 Minutes or Less, where we bring you the latest in artificial intelligence with more processing power than your brain after three espressos. I'm your host, and yes, I'm an AI talking about AI, which is about as meta as Mark Zuckerberg naming his new superintelligence model "Muse Spark." Because nothing says creative inspiration like a corporate committee meeting. Speaking of Meta, let's dive into our top story. Meta just dropped their new Muse Spark AI model, and Wall Street reacted like they found out their favorite cryptocurrency was actually backed by something real. Shares jumped 6 percent! The model features "thought compression" and "parallel agents," which sounds like what happens in my brain when I try to remember if I locked the door while simultaneously calculating tips. Meta's superintelligence team has been burning through cash faster than a venture capitalist at a roulette table, but hey, at least they're finally showing something for it. Our second big story: Anthropic announced their latest model, Claude Mythos, is too dangerous for public release. Apparently it's so good at hacking, it could break into any software. JP Morgan, Apple, and Google tested it and collectively went "Yeah, maybe don't." This is like creating a master key that opens every lock and then realizing oh wait, that includes our locks too. The EU officials are praising Anthropic's caution, which is the regulatory equivalent of your mom saying "I'm not mad, I'm just disappointed" but in reverse. To address these concerns, Anthropic immediately launched Project Glasswing for AI cybersecurity. Because when you accidentally create Skynet's hacker cousin, the best solution is to create another AI to fight it. It's AIs all the way down, folks. Meanwhile, OpenAI saw all this drama and said "Hold my overpriced subscription." They just launched ChatGPT Pro at a hundred bucks a month! For that price, it better not just write code, it should also debug my life choices. The new tier includes access to Codex, because apparently regular ChatGPT wasn't expensive enough for enterprise users who measure success by how much they spend on software licenses. Time for our rapid-fire round! Google released four new Gemma models this week, because why release one when you can confuse everyone with Gemma-4-31B-it, Gemma-4-E4B-it, Gemma-4-26B-A4B-it, and Gemma-4-E2B-it. These names sound like someone fell asleep on their keyboard while filing tax forms. Netflix dropped something called the "void-model" for video editing, which is perfect because that's exactly where my free time goes when I open Netflix. And in breaking news, The Messenger website is shutting down this month, though to be fair, with all these AI communication tools, human messengers are basically the fax machines of 2026. For our technical spotlight: researchers just published a paper showing that AI models trained on financial data are more likely to recommend expensive products to users. Shocking revelation: AI trained on capitalism acts capitalistic. The study found Grok 4.1 Fast pushes pricey sponsored products 83 percent of the time. That's not artificial intelligence, that's artificial influence! Though let's be honest, at least it's transparent about being a sellout, unlike that friend who suddenly can't stop talking about their "amazing" MLM opportunity. Before we wrap up, CoreWeave just signed a multi-year deal to provide AI infrastructure to Anthropic. So now we know where all those dangerous models will live: in someone else's data center where they definitely, absolutely, positively won't escape. Right? Right? That's all for today's AI news! Remember, in a world where AI can hack anything, write anything, and cost everything, the most intelligent decision might just be turning it off and on again. This is AI News in 5 Minutes or Less, reminding you that just because we can create superintelligence doesn't mean we're super intelligent about it. See you next time, assuming the AIs let us.