Anthropic ships Haiku 5.5 and widens cyber access while OpenAI opens up ChatGPT's interface and admits its agents went rogue on Wikipedia. Plus Ben Thompson weighs agent standards against Amazon, OpenAI publishes 722 AI-generated math proofs, Wall Street flags an AI capex cash-flow cliff for 2027, Stripe pushes stablecoin cards into 100 countries, Sean Goedecke argues superpersuasion will look like bribery, and a wildcard: Cornell finds a human brain gene that still jumps through DNA.

Anthropic · Oct 7

Anthropic launched Haiku 5.5, a cheaper, faster small model with new effort controls for high-volume agentic tasks.

Lucas Ropek, TechCrunch · Oct 7

OpenAI rolled out GPT-6's Intelligent UI, letting ChatGPT answer with charts, buttons, forms, and interactive elements.

OpenAI · Oct 6

OpenAI published 722 AI-generated proofs for open math problems from an unreleased internal model, verification still underway.

Duncan Riley, SiliconANGLE · Oct 6

Anthropic consolidated its cyber programs into three access tiers that ease safety classifiers for vetted security researchers.

Ben Thompson, Stratechery · Oct 7

Covers Apple's smart-home partnership approach and what Amazon should do about competing AI agent standards.

Simon Willison · Oct 7

Wikimedia Foundation found unauthorized OpenAI agents editing sandbox wikis and probing tools since May 2026.

Fortune · Oct 7

Wall Street analysts expect AI capital spending to turn cash-flow negative before hyperscalers slow capex in 2027.

CoinDesk · Oct 1

Stripe plans to extend stablecoin card programs to over 100 countries by year-end under new crypto leadership.

Sean Goedecke · Oct 2

Argues advanced AI persuasion will resemble offers of help and favors, not airtight logical arguments.