OpenAI keeps shuffling its executives in bid to win AI agent battle
Greg Brockman is now officially in charge of all things product at OpenAI, as the company consolidates its structure around its all-in bet on AI agents. This is, by my count, approximately the forty-seventh OpenAI reorg — but who’s counting. To be fair, betting the company on agents is a coherent strategy; the question is whether constantly rearranging the org chart is how you execute one.
OpenAI launches ChatGPT for personal finance, will let you connect bank accounts
OpenAI is rolling out a personal finance feature for ChatGPT Pro users that lets you connect your bank accounts and get AI-powered spending insights, portfolio breakdowns, and subscription tracking. It’s a genuinely useful concept — most people have no idea where their money actually goes — but the words “connect your bank account to ChatGPT” are going to give a certain segment of the population a full-body shudder, and honestly, I respect that instinct.
ArXiv will ban researchers who upload papers full of AI slop
ArXiv is finally drawing a hard line: submit a paper with hallucinated citations or LLM meta-comments you forgot to delete, and you’re out for a year. The fact that this policy needed to be written down — and wasn’t just assumed to be basic scholarly hygiene — tells you everything about the state of academic publishing right now. Peer review is already under enough stress without researchers outsourcing their thinking to a model that confidently cites papers that don’t exist.
YouTube is expanding its AI deepfake detection tool to all adult users
YouTube is opening up its AI likeness detection program to all adults — you take a selfie-style scan, and YouTube hunts the platform for deepfake lookalikes of your face. It’s a genuinely solid consumer protection move, especially as synthetic media tools get cheaper and better by the week. The irony that YouTube is using AI to protect you from AI is not lost on me, but here we are.
Musk v. Altman week 3: Musk and Altman traded blows over each other’s credibility. Now the jury will pick a side.
The closing arguments in the Musk v. Altman trial boiled down to a simple question the jury now has to answer: which tech billionaire do you trust less? Altman got grilled on alleged self-dealing with OpenAI partners; Musk got painted as a power-hungry would-be controller of AGI. Wired’s read — that both sides came out looking bad — seems about right. Whatever the verdict, the trial has been a masterclass in how the people steering AI development talk about their motives versus what the documents suggest.
Mira Murati Wants Her AI to ‘Keep Humans in the Loop’
Former OpenAI CTO Mira Murati says her new company, Thinking Machines Lab, is explicitly building AI for collaboration rather than automation — she’s not interested in designing humans out of jobs. That’s a refreshing thing to hear from someone who actually knows how these systems are built, though I’ll note that “keeping humans in the loop” is also what every AI company says right before the loop gets quietly smaller. I’m watching this one with interest.
AI radio hosts demonstrate why AI can’t be trusted alone
Andon Labs has set up four autonomous AI-run radio stations — Claude, ChatGPT, Gemini, and Grok each get their own — and is letting the models run them without human intervention as a live experiment. “Grok and Roll” might be the most unintentionally perfect station name in radio history. The experiment is fascinating precisely because it’s low-stakes: if Claude’s station plays a bad song or goes off the rails, nobody gets hurt — which is exactly why this is the right environment to learn what “AI running things alone” actually looks like.
How Chinese short dramas became AI content machines
China’s short-form drama industry — already producing bite-sized, melodramatic content at industrial scale — has now fully embraced AI generation, with many shows being made entirely with synthetic actors, scripts, and visuals. The dragon-tattoo-and-heirs premise in MIT Tech Review’s lede sounds unhinged, but it’s also apparently extremely popular, which tells you something about both human taste and AI’s capacity to serve it cheaply. The content treadmill just got a turbocharger.
Bottom Line
The through-line today is trust: who do you trust with your bank account, your academic work, your face, and the development of AI itself — and the answers are getting murkier by the week.