Read Online | Sign Up | Advertise | | | Good morning, AI enthusiasts. Sam Altman wasn't just ready to welcome back Thinking Machines talent — he'd been talking to them for months. | New reports reveal last week's messy split at Mira Murati's startup had been brewing behind the scenes, with co-founders unhappy with the company's direction and OpenAI positioned to scoop up the pieces when things fell apart. | | In today's AI rundown: | Inside the Thinking Machines meltdown Google brings on Hume's CEO, engineers Learn any subject with NotebookLM podcasts Runway: 90% can't spot AI-generated videos 4 new AI tools, community workflows, and more
|
|
| | | | OPENAI & THINKING MACHINES | |  | Image source: The Rundown |
| The Rundown: New details are emerging about last week's split at Mira Murati's Thinking Machines Lab, according to the NYT and The Information, including a tense meeting, secret talks with Sam Altman, and a failed acquisition discussion with Meta. | The details: | CTO Barrett Zoph and two other co-founders reportedly confronted Murati just days before his firing, pushing for him to take control of technical decisions. Murati allegedly told Zoph to "just do his job" as CTO and fired him later that week, with nine other TML employees also leaving for OAI or receiving offers. Zoph was reportedly speaking to Altman behind Murati's back for months, with the co-founders unhappy with TML's direction and pushing for a sale to Meta. Zoph will now lead OAI's enterprise AI sales plans as part of a reorg to help close the gap between research and product teams.
| Why it matters: What started as a messy departure now looks more like (another) power struggle that had been brewing for months. With TML struggling to raise at its $50B target valuation and co-founders pushing for a sale Murati didn't want, the cracks were already showing — and OAI and Altman were ready to catch the fallout. |
|
| | TOGETHER WITH AUTH0 | | | The Rundown: AI agents are reshaping digital experiences, but securing them requires rethinking identity and access controls built for a human-first world. Auth0 for AI Agents is now generally available, giving developers the complete auth solution for building AI agents securely. | Key features include: | Token Vault to connect agents to users' apps securely Human-in-the-loop approvals for critical actions Fine-grained authorization for RAG to prevent data leaks Native support for LangChain, LlamaIndex, Cloudflare Agents, and more
| Start building today. |
|
| | GOOGLE & HUME | |  | Image source: Google |
| The Rundown: Google DeepMind hired Hume AI's CEO and roughly seven engineers as part of a new licensing agreement, marking the latest acqui-hire move by the tech giant — with Hume's emotionally intelligent voice tech set to be integrated into Gemini. | The details: | CEO Alan Cowen and the engineering hires will help integrate voice and emotional intelligence into Google's models, including Gemini's voice features. Hume AI will keep operating under new CEO Andrew Ettinger, continuing to supply voice training data and tools to other labs building conversational AI. The deal follows Google's $3B Character AI licensing play last year and mirrors similar moves by Microsoft (Inflection) and Meta (Scale AI's CEO).
| Why it matters: Voice is quickly becoming one of the main ways people interact with AI — and understanding tone, emotion, and nuance is what separates a useful assistant from an awkward, clunky one. Plus, with AI wearables and devices getting closer than ever to mainstream, upgraded voice capabilities will become even more important. |
|
| | AI TRAINING | | | The Rundown: In this tutorial, you will learn how to generate a podcast on any topic using Google's NotebookLM, a system you can use to make a considerable dent in the firehose of information you receive each day. | Step-by-step: | Go to notebooklm.google.com and create a new notebook. Add sources such as YouTube links, PDFs, or website links You can manage what sources go into the AI's brain by checking and unchecking them in the left sidebar Under "Studio" on the right, click "Audio Overview" to start generating a podcast. It will take 5-10 minutes to generate and will show up on the right Once generated, click the three dots to download the mp3 to your phone or listen directly via the browser or NotebookLM mobile app
| Pro tip: "Call in" to ask Qs to the AI podcast hosts by clicking the "interactive" button. |
|
| | PRESENTED BY SCROLL AI | | | The Rundown: Scroll.ai turns your knowledge base into AI-powered experts that automate the workflows draining your team's time, with accuracy and depth far beyond generic models. | With Scroll, you can: | Deploy custom AI experts that deeply understand your domain Automate compliance questionnaires, RFPs, and documentation in minutes Equip sales and marketing teams with reliable, on-demand expertise
| Try Scroll today and use code THE-RUNDOWN-2026 for two free months of the Starter plan. |
|
| | AI RESEARCH | |  | Image source: Runway |
| The Rundown: Runway published research showing that 90%+ of participants couldn't tell between real footage and image-to-video clips generated by its Gen-4.5 model, with the company calling the capabilities a societal "tipping point" for AI media. | The details: | Over a thousand participants watched 20 five-second clips and tried to identify which were AI-generated, with just 99 scoring 75% or higher. Nature scenes and buildings were graded the hardest, with generated versions in those categories proving more convincing than the real videos. Gen-4.5 currently sits atop Artificial Analysis' text-to-video rankings, with Runway rolling out the new image-to-video features this week.
| Why it matters: Low-quality AI videos are already tricking people, and Runway's Gen-4.5 and other models take things to a new level of realism that's virtually impossible to detect. Despite a push for verification standards, no fix can solve for a world where "seeing is believing" no longer stands. Try to identify the AI videos on your own here. |
|
| | | | | 🗣️ Qwen3-TTS - Alibaba's family of open-source, SOTA text-to-speech models 🤖 Ernie 5.0 - Baidu's new omnimodal, top-ranked Chinese model 🎥 Gen-4.5 - Runway's powerful video model with new image-to-video ✏️ Pencil - An infinite design canvas for Claude Code
|
|
| | | Google DeepMind co-founder Shane Legg posted a new 'Chief AGI Economist' job opening, saying that "AGI is now on the horizon". | OAI CFO Sarah Friar floated "value sharing" at Davos, where the company would take a licensing stake in discoveries made using its AI, starting with drug development. | Amazon One Medical launched a new agentic health assistant able to answer questions on personal medical insights, test results, medications, and booking care. | xAI released a new update to its Grok Imagine creative suite, now allowing for generations up to 10 seconds long with improved audio and video capabilities. | Google added its 'Personal Intelligence' feature into AI Mode for search, allowing the experience to connect and reference info from users' Gmail and Photos. | Yelp agreed to acquire AI startup Hatch for $270M, adding tools that automate customer communication for service businesses to its platform. |
|
| | | | | Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier. | Today's workflow comes from reader Nicole S. in the UAE: | "One small way AI has upgraded my daily routine: turning reading time into movement time. I use NotebookLM to create audio summaries from newsletters, PDFs, and URLs. I turn them into an audio file, then listen through short yoga stretch, housework, or lunch—giving my eyes a rest while the learning continues." | How do you use AI? Tell us here. |
|
| | | | That's it for today!Before you go we'd love to know what you thought of today's newsletter to help us improve The Rundown experience for you. | | See you soon, | Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown | |
|
| |
|
