June 4 brought a wave of new frontier models plus a reminder of how much money and risk now rides on them. Microsoft shipped seven new MAI models including a coding-focused MAI-Code-1-Flash baked into GitHub Copilot, NVIDIA opened Cosmos 3 for physical AI and teased Nemotron 3 Ultra, and Anthropic's Opus 4.8 reportedly broke the ARC-AGI-3 benchmark while drawing fresh scrutiny over model welfare. On the business side, Anthropic filed a confidential draft S-1 for what could be one of the largest tech IPOs ever, Alphabet said it will raise $80 billion in stock sales to fund AI infrastructure, and AWS added GPT-5.5/5.4 plus Codex to Bedrock. Meanwhile security researchers flagged a steady drip of AI-tooling supply-chain risk, from a GitHub-Issue-based poisoning attack on Claude Code to a one-click VS Code token-stealing bug, and Wix cut 1,000 jobs. For engineers the throughline is that model quality, distribution, and attack surface are all scaling together, faster than most teams' review processes can keep up.

• Microsoft ships seven MAI models incl. coding-focused MAI-Code-1-Flash in Copilot
• NVIDIA opens Cosmos 3 for physical AI, teases Nemotron 3 Ultra
• Opus 4.8 reportedly breaks ARC-AGI-3 amid model welfare debate
• Anthropic files confidential draft S-1; Alphabet raises $80B for AI buildout
• Agentic coding tools become live attack surface: Claude Code, VS Code both hit