Story cards stay below the sticky dock while audio, chapters, date, and brief navigation remain accessible.
Anthropic confirmed that over 80% of code merged into its internal repositories is now written by Claude — a remarkable self-referential milestone. Simultaneously, the company expanded Project Glasswing (Claude Mythos Preview) to roughly 150 partner organizations, where it has already flagged more than 10,000 high- and critical-severity security vulnerabilities. AI writing AI infrastructure is no longer hypothetical.
Microsoft unveiled MAI-Code-1-Flash, its first in-house code-generation model claiming 10x cost efficiency over GPT-5.5, alongside MAI-Thinking-1, its first reasoning model — signaling real intent to reduce OpenAI dependence. Windows is being repositioned as an OS for AI agents via Windows 365 for Agents, Aion 1.0, and new IQ context layers (Microsoft IQ, Work IQ, Web IQ). The Pentagon simultaneously signed a $9.69 billion enterprise deal covering Azure and Copilot — Microsoft's largest single government contract ever.
Google released Gemini 3.5 Flash into general availability — frontier-class intelligence at 4x the speed of comparable models, priced at $1.50 input / $9 output per million tokens with a 1M-token context window. Google DeepMind also dropped Gemma 4 12B, an encoder-free multimodal model with native audio input that runs on a consumer laptop with 16GB of RAM. The local-capable frontier model category is accelerating fast.
Anthropic submitted a draft Form S-1 to the SEC, officially starting the clock on a public offering that could set records. The filing follows a $65 billion Series H round that pushed Anthropic's valuation to $965 billion — surpassing OpenAI at $852 billion to become the world's most valuable AI startup. The IPO would be a defining moment for AI as a public-market asset class.
Demis Hassabis made a pointed public statement on June 4 arguing that humanity needs to begin preparing immediately for the "next stage of AI development," with AGI on track for 2030. This landed alongside DeepMind's acquisition of 20-plus Contextual AI researchers in a deal valued at $80–90 million. When the CEO of the lab broadly acknowledged as a frontrunner sets a four-year AGI clock, the industry listens.
OpenAI updated GPT-Rosalind with stronger agentic coding, drug-discovery benchmarks, genomics performance, and bioinformatics plugin support — deepening its push into one of AI's highest-value verticals. ChatGPT also added Active Sessions security controls, live job-search and resume tools, and expanded Codex with Computer Use on Windows. OpenAI's frontier models and Codex are now available directly on AWS.
Meta unveiled the Meta Business Agent under its new "Meta One" subscription brand, letting businesses deploy AI agents across WhatsApp, Messenger, and Instagram for customer service and sales — with over a million businesses already on earlier versions. The same week, Wired uncovered unreleased code in Meta's AI app for a facial recognition feature codenamed "NameTag." Scale of reach plus facial recognition in waiting is a combination that will draw scrutiny.
President Trump signed a sweeping AI executive order directing agencies to accelerate AI adoption, harden cybersecurity, and protect critical infrastructure — framed explicitly around maintaining U.S. global AI leadership. The order lets AI labs provide models to the government for testing up to 30 days before public release, down from a proposed 90 days after industry pushback. Washington's tone has shifted decisively from guardrails to acceleration.
Bipartisan House lawmakers dropped a draft national AI framework on June 4 aimed at preempting a patchwork of state laws — requiring frontier developers to publish safety protocols, mandating incident reporting, and codifying the Center of AI Standards and Innovation. It is Congress's first formal legislative response to the White House AI agenda, and its bipartisan authorship gives it unusual runway. Vermont banning therapy bots and California advancing 30 AI bills underscore exactly why a federal framework has urgency.
Mistral shipped Vibe, a unified agentic platform that retires Le Chat and introduces persistent memory, Work Mode for long-horizon tasks, and Code Mode for remote coding agents. The company also released its open-source Search Toolkit in public preview and announced a new 10-megawatt inference facility in Les Ulis, France opening Q3 2026. Mistral is clearly positioning itself as a credible European alternative with serious enterprise intent.
