Pulse Daily — Saturday, August 22, 2026

On Saturday, August 22, 2026, the AI landscape surged with Anthropic eyeing a historic IPO, Nvidia in acquisition talks with chip startup Rebellions, a critical security flaw actively exploited in the widely-used Ray AI framework, Google's Gemini 3.7 Flash rewriting coding benchmarks at half the price, a new benchmark exposing frontier LLMs' failure at scientific hypothesis generation, Cloudflare launching agent-native browser infrastructure, the EU's AI identification rules taking hold, OpenAI's Astra having cracked open mathematics problems, the agent harness rankings reshuffling, Las Vegas opening its streets to robotaxis, and a government-switched-off AI model raising governance alarms. Stories: Anthropic Is Preparing for an IPO — and It Could Be a Large One · Nvidia Moves on a Model-Building Startup — Chipmakers Stack Fresh AI Deals · SK Hynix Launches $28.6B Buyback After AI-Capex Scare · A Botnet Weaponized the Ray AI Framework Two Days Before Anyone Knew It · Gemini 3.7 Flash Jumps 16 Coding Points in Three Weeks — Not by Getting Bigger · Frontier Models Score 3–15% on New Hypothesis Benchmark · Cloudflare Gives AI Agents a Lean Browser and a Way to Pay · Claude Code Tops August AI Coding-Agent Rankings — Cursor Leads In-Editor · A Government Switched Off an AI Model — and the Rules Are Still Being Written.