The AI world spent July 30 arguing over openness and access even as the infrastructure buildout keeps accelerating: tech companies are now sinking roughly $1 trillion into AI infrastructure this year alone, Anthropic pushed back on claims it wants open-weight models banned, and Moonshot's Kimi K3 landed with a 256k context window as a serious open-weight contender. On the security side, a ChatGPT Workspace Agent Builder flaw let attackers manipulate initialization state via URL parameters, and WIRED flagged that private Claude chats were showing up in Google and Bing search results, both reminders that agent and chat surfaces are shipping faster than their guardrails. For engineers, there's also a systems-programming thread worth a look, from a RISC-V emulation deep dive to TanStack's public rethink of React Server Components.

• Tech industry AI infrastructure spend hits roughly $1 trillion this year
• Anthropic denies wanting open-weight models banned, backs disclosure norms instead
• Kimi K3 ships with a 256k context window as an open-weight option
• ChatGPT Agent Builder could be manipulated via URL init parameters
• Private Claude chat transcripts turned up indexed in Google and Bing search