Anthropic drops Mythos — too powerful for public release. OpenAI ships a super app. $300B in Q1 VC. EY puts 130,000 auditors on agents. Eight stories from the week AI stopped asking for permission.
Anthropic just released a model it says is too dangerous for the public — and handed it to Apple, Microsoft, and Amazon to defend critical infrastructure. OpenAI merged ChatGPT, Codex, and a browser into one super app with 900 million weekly users. Venture capitalists poured $300 billion into startups in a single quarter — 81% of it into AI. EY deployed multi-agent systems across 130,000 auditors in 150 countries. And a Tufts research team proved you can slash AI energy consumption 100× by teaching robots to think more like humans. Eight stories from the week AI stopped asking for permission.
Anthropic Says Its New Model Is Too Powerful to Release.
So It Gave It to Apple, Microsoft, and Amazon Instead.
Anthropic officially launched Claude Mythos Preview on April 7 — the model first exposed through a CMS misconfiguration on March 26 that made 3,000 unpublished assets publicly searchable. The benchmarks justify the drama: 93.9% on SWE-bench Verified (up from 80.8% for Opus), 97.6% on USAMO, and 83.1% on CyberGym versus 66.6% for its predecessor. Rather than a public release, Anthropic created Project Glasswing — a defensive cybersecurity initiative giving 12 organizations including Apple, Microsoft, Amazon, CrowdStrike, and Palo Alto Networks access to the model for securing critical infrastructure. Forty additional organizations received access alongside $100 million in usage credits.
Anthropic just invented a new distribution model: restrict the most powerful AI to defensive applications and let the security community validate it before the public gets near it. If Mythos performs as advertised, the gap between frontier capability and public access just became a deliberate strategic choice — not a lag.
OpenAI Merged ChatGPT, Codex, and a Browser Into One App.
900 Million Users Just Got a Desktop Operating System.
OpenAI launched ChatGPT 5.5 on April 6, unifying its chatbot, Codex coding agent, and Atlas browser into a single desktop application. The integration allows users to switch between writing, coding, and web browsing without leaving the window. App integrations went live simultaneously — Spotify for AI-generated playlists, Uber Eats for restaurant discovery, Wix for one-prompt website creation, and Google Drive for unified Docs, Sheets, and Slides access. With 900 million weekly active users and $25 billion in annualized revenue, OpenAI is no longer building a chatbot. It's building an operating system that happens to understand natural language.
This is OpenAI's WeChat moment. The super app model — chat, commerce, tools, browsing in one surface — worked in Asia because it owned the user's attention. If ChatGPT becomes the default desktop for 900 million people, every SaaS company becomes an integration, not a destination.
Investors Poured $300 Billion Into Startups in One Quarter.
The first quarter of 2026 was the biggest in venture capital history. Investors deployed $300 billion across 6,000 startups globally — up 150% year over year. AI companies absorbed $242 billion, representing 81% of total global VC, up from 55% in Q1 2025. The concentration is staggering: four companies — OpenAI ($122B), Anthropic ($30B), xAI ($20B), and Waymo ($16B) — captured $188 billion, roughly 65% of all global venture capital. U.S.-based companies raised 83% of the total. The quarter alone accounted for nearly 70% of all venture capital invested throughout 2025.
When four companies absorb two-thirds of global venture capital, that's not a market — it's a power law with a pulse. The real question: what happens to the 5,996 other startups competing for the remaining 35%? The AI boom is real, but the capital concentration suggests most of the value will accrue to a handful of winners.
EY Just Deployed AI Agents Across 130,000 Auditors.
The Big Four's Agentic Transformation Has Begun.
EY launched a global multi-agent AI framework on April 7, embedded directly in its Canvas audit platform — the system that processes 1.4 trillion lines of journal entry data per year across 160,000 audit engagements in 150 countries. The framework starts with a core assistant plus three specialized agents covering roughly 20 modular capabilities, integrated with Microsoft Azure, Foundry, and Fabric. EY's target: 100% agent-supported audit activities by 2028. The firm is part of Microsoft's Frontier Firm AI Initiative, one of only 14 organizations deploying advanced AI at scale. Junior staff will face a steep learning curve as agents handle tasks that once defined the first three years of an audit career.
EY just turned professional services into an agent deployment story. If 130,000 auditors can run multi-agent workflows by 2028, the billable-hour model doesn't survive contact with this technology. Watch for Deloitte, PwC, and KPMG to follow within months — or get disrupted by firms that already did.
Tufts Researchers Cut AI Energy Use 100× —
By Teaching Robots to Think Like Humans.