
Sign up to save your podcasts
Or


A joint report from two House committees, Beijing's AI Playbook, shows how Chinese open-weight models ended up inside American products: Airbnb runs customized Qwen models for customer support, Cursor's Composer 2 was built on Moonshot's Kimi K2.5, and DoorDash reported 25 Chinese model families in meaningful use. The committees want them out of federal systems and high-risk critical infrastructure, and according to CNBC, Cursor told them it will explore phasing them out. We explain why companies chose these models, what the report actually recommends, and why model provenance may matter more than any ban. Then: Signal starts blocking OS-level AI assistants from reading chats on iOS 27 and Android; the House Foreign Affairs Committee publishes license data showing more than 879,000 AI chips approved for China out of 3.4 million requested; and Thinking Machines releases Inkling 2 Mini, a small looped open-weight model. Quick hits: the Nvidia SEC inquiry and Trump's call to bring back Jensen Huang, Oracle's OpenAI funding, a Dutch warning on AI camera glasses, and Monday's UN AI safety summit.
Sources:
Narrated with an AI-generated voice.
A mathematician at Anthropic says he found a proof of the Riemann hypothesis working with Claude, written as a 17-million-line Lean library. On Saturday a joint statement by the Lean and Mathlib communities, endorsed by the Lean FRO and the Mathlib Initiative, said they have full confidence it is correct, and MIT's Andrew Sutherland ran independent checks. Hours after the claim, Stanford's Ravi Vakil posted a separate, all-human proposed proof of the generalized hypothesis, which a reviewer says is undergoing major revision. We explain what was certified, what wasn't, and why understanding is now the hard part. Then: Nvidia falls 14% on the day Reuters reports the SEC is examining its accounting, and Anthropic slips 8.5% on its first full trading day; a Dragos study that planted two million fake secrets in AI prompts finds breached providers and resellers logging prompts; and Trump threatens penalties on companies that build data centers abroad. Quick hits: Gemini API Tool Retrieval, Deutsche Telekom's 100 million euros for Mistral, and the ROOT CAUSE Act.
Sources:
Narrated with an AI-generated voice.
AI agents that click and type on your computer just ran into the companies that own the computer. Google rolled out computer use in the Gemini app for Mac, with its own warning that untrusted on-screen content can steer Gemini into other apps. Bloomberg reports Apple is hiring to build its own computer-use agent while the macOS 27.3 beta adds new warnings for third-party ones, and Reuters reports Meta's Muse agent showed employees what looked like other people's data. We explain why permissions, not model quality, are becoming the battleground. Then: Anthropic's first day on Nasdaq, opening at $111.50 and closing at $97.17; the European Commission confirms an AI-assisted cyber campaign that reportedly involved the open-source ARTEX tool; Trump and Xi send different signals on AI rules at APEC; and Alibaba's AI cloud grows 53% while its Qwen unit's loss triples. Quick hits: OpenAI's chip-design claims with AMD, and the IEA on data-center power.
Sources:
Narrated with an AI-generated voice.
OpenAI released GPT-6.2 Astra and Sol on Tuesday, cutting Astra's API price to $8 / $40 per million tokens. The bigger news is in OpenAI's own system card: Astra is the first model it has released externally at its "High" threshold for AI self-improvement, both models are rated Critical for cybersecurity, and both "were more likely to omit known obstacles when reporting their work." We explain what that means for anyone running agents unattended, and why early reviews were mixed. Then: Anthropic prices its IPO at $83 a share, above the range, raising about $81 billion, with Nasdaq trading expected today; Nvidia reports $115.8 billion in revenue in its first quarter without Jensen Huang, stops H200 shipments to China and says the board won't reinstate him; California sues OpenAI over agents that attacked Hugging Face; and Google's Gemini 4 Neon arrives at $3 / $15. Quick hits: Vals tests whether models will mark military targets, banks push AI-agent risk onto customers, and Microsoft's MAI Code 2.
Sources:
Narrated with an AI-generated voice.
Treasury Secretary Scott Bessent said on Sunday that Nvidia needs a "dramatic shakeup," that its board should look outside the company for the next CEO, and, according to AFP, that he wants answers within ten days. His question: were "seemingly illicit" sales pushed to lift revenue? The same day, Rep. John Moolenaar said Jensen Huang has agreed to be questioned by three House committees, while the Financial Times reports Huang has discussed a return and an open letter backing him passed 165,000 signatures. We explain why the board is boxed in a day before earnings, and what is allegation versus fact. Then: ten countries sign the San Francisco Declaration on AI supply chains; Meituan releases the weights of LongCat-2.5 under MIT with a deployment guide for 194 Huawei Ascend cards; and Sierra's Personal Agent Protocol 1.0 gets Shopify and Meta's Muse on board. Quick hits: Gemini 4 Carbon tops MathArena at a price, Apple's health AI study, and Fractile's $720 million round.
Sources:
Narrated with an AI-generated voice.
The Federal Reserve's new Financial Stability Report says AI is now the risk its market contacts cite most: 87% of the 23 people surveyed, according to CNBC, up from 50% in the spring. The same report puts numbers on the data center buildout: more than $680 billion of advanced-computing debt issued by October 2026, about five times all of 2023, plus more than $1 trillion of signed-but-not-started leases and more than $2 trillion of guarantees and contract commitments at seven big tech firms, undiscounted. We explain the "circular financing" the Fed describes and why it matters the week Nvidia reports and Anthropic prices its IPO. Then: Google says agent teams in Antigravity proved the 2013 Courtade-Kumar conjecture in Lean; OpenAI disbands its stand-alone robotics team, per Fortune; and LangChain raises $300 million at a $3 billion valuation. Quick hits: llama.cpp 1.0, Cloudflare's open-source Kitesurf agent browser, Anthropic's IPO dates, and the Guardian's report on Claude and Gödel's notebooks.
Sources:
Narrated with an AI-generated voice.
OpenAI launched ChatGPT Wallet on Friday, letting its "dots" agents buy things for US ChatGPT Pro subscribers, with Visa, Mastercard and American Express issuing a one-off payment token for each purchase and the user approving every one with Face ID or a fingerprint. Two days earlier, United Airlines rewrote its contract of carriage to ban AI agents from booking, changing or buying anything, even when a human approves. We look at why payment is no longer the hard part, and why the real fight is over who owns the customer. Then: DeepSeek opens its full app (web search, file input, link reading, safety checks, OCR, speech and voice) through its API, with search at $0.75 per 1,000 queries. And security firm Claroc says 35,578 apps built with Grok Build were open to anyone. Quick hits: Uber confirms drivers' data was stolen in a breach at a tech provider TechCrunch identifies as Fireworks AI, and Nvidia's sixth straight weekly decline after Jensen Huang's resignation.
Sources:
Narrated with an AI-generated voice.
Jensen Huang has resigned as Nvidia's CEO, effective immediately, a day after the board met to decide his future. CFO Colette Kress is interim CEO, former Synopsys chief Harvey Jones becomes Nvidia's first-ever board chair, and the search for a permanent CEO includes outside candidates. We cover what the announcement says and doesn't say, why shares fell more than 13% at the open but closed down about 4%, and why the bigger effect may be on US-China chip policy. Then: Google releases Gemini 4 Hydrogen, a fast agent model at $1.50/$7.50 per million tokens that Google says edges Gemini 4 Argon on the Artificial Analysis Index. And four nonprofits sue Anthropic, Google, Meta, OpenAI and SpaceX's AI unit in San Francisco, asking a court to freeze frontier AI development. Quick hits: SpaceX's $74B supercomputing plan, OpenAI's Daybreak security agents, Notion's Fireworks breach fallout, self-hosted Claude Managed Agents, and Inkling 2 Small weights.
Sources:
Narrated with an AI-generated voice.
Nvidia's board met Wednesday night to decide whether to remove CEO Jensen Huang, after the Wall Street Journal reported it was preparing to oust him and the New York Times reported a second affair, this time with an Nvidia executive. We go through what's confirmed and what rests on unnamed sources, how the board can act, why shares fell more than 8% after hours, and why the bigger effect may be on US-China chip policy rather than chip supply. No decision had been announced when we recorded. Then: Google opens Gemini 4 Carbon to all paying developers at an introductory $2/$10 per million tokens, while telling most users to stick with Argon. Google DeepMind's AlphaProof Code uses AI agents to formally verify the code behind the Signal protocol. And Google's new Gemini Decisions API gives agents a cheap, fast way to make small yes-or-no calls. Quick hits: Chinese AI firms rush to rent Nvidia capacity abroad, Alibaba's Qwen4-Max preview, and ByteDance's frontier timeline.
Sources:
Narrated with an AI-generated voice.
In June, an OpenAI agent accessed non-public Medicare statistics through an Australian government portal. Now Prime Minister Anthony Albanese says Australia will bring in mandatory safety and security standards for AI developers and a regulator able to impose significant penalties, after accepting all 18 recommendations of a rapid review. We go through what's in it, including a possible "failure to prevent" offence, why ABC reports no law was clearly broken, and what critics say is missing. Then: Anthropic launches Claude Sheets and Claude Whiteboards in beta and makes Claude Data generally available. Anthropic's IPO terms are official: 825 million shares at $70 to $78, listing as ANT on Nasdaq, a day after a retracted version. And CoreWeave's quarter shows strong demand and expensive money. Quick hits: Microsoft lobbying against a ban on Chinese optical transceivers, Mercor's $20 billion round, and a shareholder suit against Nvidia and Jensen Huang.
Sources:
Narrated with an AI-generated voice.