Pulse Daily — Sunday, October 18, 2026

On Sunday, October 18, 2026, AI's dual nature crystallized across eleven major stories: Amazon's Atlas tool secretly scored UK warehouses for union risk, California finalized a kill-switch executive order, DeepSeek dropped a stealth model update beating rivals at a fraction of the cost, major US banks blocked unauthorized AI agents from customer accounts, StepFun open-sourced a 600-billion-parameter model, Claude Mythos 5 redesigned quantum circuits up to 17 times more efficiently, Utah's AI prescription pilot expanded, Anthropic's IPO valuation faced sharp investor skepticism, Jefferies warned AI earnings growth could crash by 2028, AI data center power infrastructure reached a crisis point with waitlists stretching to 2031, and a robotics firm signed a landmark deal with DR Horton to build American homes. Stories: Amazon's Secret Union-Watch AI Just Showed Up in Court · The Governor Who Vetoed an AI Safety Bill Just Ordered His Own Version · DeepSeek Pushed Out a Faster, Leaner Model Overnight · China's StepFun Just Open-Sourced a 600-Billion-Parameter Model Built for Long Context · Anthropic's AI Rewrote Quantum Computing Blueprints — Using 17 Times Fewer Gates · Top Investors Now Say Anthropic Should Take Less Money at a Much Lower Price · Jefferies Tells Investors to Get Out of AI Stocks — Earnings Growth Is Fading · Two-Thirds of America's Planned AI Data Centres May Never Get Built, Bernstein Says · The Company Betting $50 Billion on AI Data Centres Just Had Its Worst Week · Schwab Is Blocking Dozens of Rogue AI Agents From Client Accounts Every Week · Agents Need a Passport — and Some Are Already Traveling Without One · Utah Just Let an AI App Write Prescriptions — a First for Any US State · Sequoia Just Bet Over $500 Million That Robots Need Human Motion Tutors.