Anthropic filed a confidential draft S-1 for what could be the largest tech IPO on record, while Alphabet said it will raise $80 billion in stock sales to fund AI infrastructure, underscoring just how capital-intensive the frontier race has become. On the model side, Claude Opus 4.8 keeps generating headlines - a system card, a 'model welfare' essay, and a reported ARC-AGI-3 breakthrough - alongside NVIDIA's Cosmos 3 physical-AI model and Nemotron 3 Ultra, plus MiniMax M3 landing on Ollama Cloud. Meanwhile security researchers flagged a wave of attacks aimed squarely at AI coding tools: a GitHub-issue supply-chain poisoning technique against Claude Code, fake Anthropic sites dropping infostealers, and a DOM-based exfiltration tool called LLMReaper. For engineers, the throughline is that agentic coding tools are now both the productivity unlock and the newest attack surface, and permissions - not model capability - are becoming the real bottleneck.

• Anthropic files confidential draft S-1, teeing up a potential mega IPO
• Alphabet to raise $80B in stock sales to fund AI infrastructure buildout
• Claude Opus 4.8 reportedly breaks ARC-AGI-3 and gets a full system card
• NVIDIA ships Cosmos 3 physical-AI model and Nemotron 3 Ultra
• Attackers are poisoning Claude Code's supply chain via GitHub issues