The Atlantic Built a Searchable Database of Music Scraped for AI Training
Reporter Alex Reisner has done the music industry a considerable favor — and possibly the AI industry a considerable headache — by uncovering four datasets totaling over 21 million tracks used to train AI models and making them fully searchable by the public. Two of those datasets alone clock in at 12 million and 9 million tracks, which is a polite way of saying “basically all of music.” At this point the question isn’t whether the AI industry scraped copyrighted music; it’s whether anyone is going to do anything about it before the lawyers finish their coffee.
Nobel Laureate John Jumper Is Leaving DeepMind for Anthropic
John Jumper — co-winner of the 2024 Nobel Prize in Chemistry for his work on AlphaFold — is departing Google DeepMind for Anthropic, and TechCrunch notes he’s not the only big name walking out the door. When you’re losing Nobel laureates to the competition, the retention problem has officially moved from “HR issue” to “existential question.” Google has the compute, the data, and the cash, and yet talent keeps finding reasons to leave — that’s a story worth watching closely.
Siri AI Hands On: A Smart, Helpful Assistant
Wired’s hands-on with the new Siri describes it as conversational, omnipresent, and — brace yourself — actually helpful, which is doing a lot of heavy lifting given Siri’s previous decade of enthusiastic mediocrity. Meanwhile, The Register’s take from earlier this month lands in roughly the opposite galaxy, calling the new implementation a “cumbersome mess” that killed Spotlight and replaced it with force-fed Apple Intelligence bloat. The truth is probably somewhere in between, which means Apple has achieved “fine” — the most expensive bar in consumer tech history.
Signal’s Meredith Whittaker Wants You to Remember That AI Chatbots ‘Are Not Your Friends’
Signal president Meredith Whittaker delivered the bluntest PR-free message of the week: “These are not your friends. These are not conscious beings. These are not sentient interlocutors.” It’s a shot across the bow at the entire companion-AI industry, which has spent considerable money and effort making you feel exactly the opposite. She’s right, of course, but try telling that to the millions of people who find their AI chat more emotionally consistent than any human in their lives — that’s the uncomfortable part of this conversation nobody wants to have.
Meta’s AI Workers Are Revolting
Wired’s Uncanny Valley podcast digs into the dysfunction inside Meta’s newly formed AI unit, where already-low employee morale has been driven even further into the ground by internal chaos. Zuckerberg has been dangling $100 million offers to poach top AI talent from competitors while apparently not noticing the engineers already inside the building are miserable — which is, frankly, a very on-brand way to run a company. Building AGI on a foundation of demoralized employees is a bold strategy.
Salesforce’s Internal AI Leaderboard Has Teams Competing for Little Trophies
404 Media got a look at Salesforce’s internal AI adoption leaderboard, which ranks teams by executive and includes a feature highlighting employees who haven’t earned AI badges yet — helpfully labeled “click to see who 👀.” Nothing says “we’re transforming our culture around AI” quite like a passive-aggressive surveillance dashboard that publicly names the holdouts. Marc Benioff has apparently decided that gamification and mild social shame are the secret ingredients to enterprise AI adoption.
Who Decides When AI Is Too Dangerous?
The Verge’s Decoder podcast takes on the big question that nobody in Washington seems eager to actually answer: who has the authority to say a model is too dangerous to deploy? With Anthropic’s Claude, the Trump administration, and Pentagon AI procurement all colliding in the same news cycle, it’s clear the governance frameworks are years behind the technology. The uncomfortable reality is that right now, the companies themselves are largely making these calls — which is a bit like asking the restaurant to write its own health inspection.
Railway Secures $100M to Challenge AWS With AI-Native Cloud
Railway, a cloud platform that somehow acquired two million developers without spending a single dollar on marketing, just closed a $100 million Series B to take on AWS by building infrastructure designed from the ground up for AI workloads. The pitch is that legacy cloud architecture wasn’t built for the inference-heavy, latency-sensitive demands of modern AI apps — and they’re not wrong. Whether “AI-native cloud” is a real product category or a fundraising deck category, we’re about to find out.
Bottom Line
The talent wars, the copyright wars, and the governance wars are all accelerating at once — and nobody’s waiting for the referees to show up.