Anthropic unveils Mythos, their new frontier model launching with Project Glasswing—a cybersecurity initiative that's already uncovered thousands of zero-day bugs. Meanwhile, OpenClaw faces a critical 9.8 severity vulnerability with 63% of instances running wide open. Plus, Intel joins Musk's Terafab semiconductor megaproject targeting one terawatt of AI compute annually in Texas.

Data centers are the new oil refineries as Iran threatens OpenAI's Stargate facility with strikes. OpenAI proposes robot taxes and 4-day workweeks in a bold economic manifesto. OpenAI alumni launch Zero Shot Capital with $100M. And Google sneaks apps into the App Store. The physical security of AI just became a national priority.

SpaceX files for IPO at $1.75T valuation with orbital data centers as growth driver. Japan commits $6.3B to capture 30% of physical AI market by 2040 amid 14th year of population decline. Anthropic publishes research on LLM emotional states. Plus Microsoft fine print drama and more.

This week: Anthropic's controversial move to charge Claude Code users for third-party harnesses like OpenClaw sparks accusations of exploiting open-source, Netflix releases VOID—an AI that erases objects from video with stunning realism, and a new Generative World Renderer lets you restyle game environments using text prompts. The battle between open and proprietary AI intensifies.

This week: Anthropic acquires Coefficient Bio for $400M to boost its life sciences ambitions, files for a new political action committee ahead of midterms, and OpenAI undergoes major executive reshuffling with key leaders moving to new roles or taking leave. Plus, what it all means for the future of AI regulation and drug discovery.

Google DeepMind releases Gemma 4, a fully open-source multimodal model family that runs frontier-level AI on smartphones. Microsoft unveils three new foundational models positioned as budget alternatives to competitors. OpenAI makes its first media acquisition with TBPN, raising questions about narrative control ahead of their IPO.

Today's episode covers three major stories: H Company's Holo3 sets a new benchmark record with only 10B active parameters while outperforming GPT-5, Anthropic's DMCA takedown gone wrong that accidentally deleted 8,100 GitHub repositories, and Cognichip's $60M raise to use AI for semiconductor chip design. The era of efficient, specialized models is here.

OpenAI closes a staggering $122 billion funding round, valuing the company at $852 billion with IPO signals flashing. Anthropic suffers two major leaks in one week, exposing internal files and 512,000 lines of source code. Plus, AI recruiting startup Mercor confirms a cyberattack linked to supply chain compromise. The wildest part? It's all real.

South Korean chip startup Rebellions raised $400M and launched new inference products as the industry shifts focus from training to inference. Mistral secured $830M in debt financing to build European data centers, while Starcloud raised $170M at unicorn valuation to launch GPU-equipped satellites into orbit. The massive capital flowing into AI infrastructure shows the industry is no longer just about bigger models—it's about making the whole stack actually work.

OpenAI has shut down Sora after burning roughly $1 million per day with fewer than 500,000 users, killing a $1 billion Disney partnership in the process. Meanwhile, Apple announces Extensions—an AI marketplace inside Siri coming in iOS 27, with Google's Gemini reportedly powering core intelligence. Plus, ByteDance delays Seedance 2.0 launch as AI video hits reality check.