White House Slows GPT-5.6, OpenAI Chips Away at Nvidia's Monopoly
Trump told OpenAI to sit on its newest model, Claude is stealing ChatGPT's paying customers, and video games are now AI training data worth $2.3B.
Subscribe free →The White House Just Became an AI Gatekeeper
OpenAI's newest model, GPT-5.6, won't be getting a public launch — the Trump administration asked the company to slow-roll it, releasing it only to select partners instead. This is reportedly framed around safety concerns, making it the clearest example yet of the executive branch directly intervening in an AI product release timeline.
Why it matters: Governments shaping when and how frontier AI models reach the public is no longer hypothetical — it's happening right now, and it sets a precedent every lab is watching.
Anthropic's Mythos Crisis Is Getting Messier by the Day
Two weeks after the Trump administration forced Anthropic to take its Mythos-class models offline with a Friday-evening ultimatum, the company still has no resolution. Anthropic dispatched executives to DC immediately but has gone radio silent since, declining to comment repeatedly while negotiations drag on with no end in sight.
Why it matters: One administration holding an entire model family hostage — with no public timeline or criteria for resolution — is a new and alarming form of AI governance by threat.
OpenAI's "Jalapeño" Chip Is Big Tech's Clearest Break from Nvidia
OpenAI has unveiled plans for Jalapeño, a custom inference chip built in partnership with Broadcom. It joins Google, Apple, and SpaceX in a growing cohort of companies engineering their way out of total Nvidia dependency — a supplier relationship that's been as expensive as it's been unavoidable.
Why it matters: When the biggest AI spender on the planet starts building its own silicon, Nvidia's lock on AI compute starts looking a lot less permanent.
Claude Is Quietly Stealing ChatGPT's Paying Customers
Data shows that consumers willing to pay for AI are increasingly choosing Anthropic's Claude over ChatGPT, even as OpenAI maintains a massive lead in total users. The paid tier is where margins live, where power users concentrate, and where platform loyalty gets forged — and Claude is winning it.
Why it matters: Free-tier headcounts are vanity metrics; paid subscribers are the business, and Anthropic is taking that ground from OpenAI one credit card at a time.
A $2.3B Bet That Video Games Can Teach AI to Think Like Humans
General Intuition has raised $320M toward a $2.3B valuation, building AI agents trained on millions of hours of gameplay footage. The thesis: the dense, moment-to-moment action data generated by games can develop something closer to genuine intuition in AI systems than traditional training pipelines.
Why it matters: If gameplay data can transfer to real-world agent performance, it opens an enormous new synthetic data pipeline at a time when quality training data is the scarcest resource in AI.
Patronus AI Raises $50M to Build Crash-Test Dummies for AI Agents
Patronus AI, founded by former Meta AI researchers, just closed a $50M round to build simulated "digital worlds" that stress-test AI agents before they're deployed in the real one. Demand is described by investors as nearly insatiable — unsurprising given how badly the industry needs reliable agent evaluation tools.
Why it matters: Agentic AI is only as trustworthy as the testing infrastructure behind it, and right now that infrastructure barely exists — Patronus is betting it can own that layer.
Quick Hits
- →Notion is shutting down its email app because most users already handed inbox control to AI agents — a rare product kill where the stated reason is "agents won" (Ars Technica)
- →TechCrunch argues the Anthropic vs. OpenAI framing is obsolete — AI capabilities now have geopolitical consequences that require collective industry action, not competition (TechCrunch)
- →Xprize founder Peter Diamandis joined the growing chorus of tech leaders arguing global surveillance improves human behavior — a take that's aging poorly (TechCrunch)
- →SAP is restructuring its commerce data architecture to support AI personalization at enterprise scale — unglamorous infrastructure work that determines whether AI promises actually ship (AI News)
AI TLDR