Gemini gets a real Buy button — Google's UCP checkout goes live
Google's first live agentic-checkout test lands in India via UCP. Plus NYC's kill-switch bills, Claude Code model locks, and OpenAI's open mental-health eval.
Copy markdown
Gemini's Buy button goes live via an open protocol
Google is testing a real “Buy” button inside Gemini and AI Mode, starting with Flipkart in India and expanding in October for the festive season. It's the first live deployment of Google's Universal Commerce Protocol (UCP), an open standard that lets agents browse and check out — merchants can keep their own checkout or use Google's hosted flow. Agentic commerce just moved from demo to an integrable surface you can build against.
NYC wants a kill switch on every AI system
NYC Council Speaker Menin introduced a 10-bill package on Sept 25 mandating kill switches, 24-hour incident reporting, third-party validation, and whistleblower bounties paid from fines — with penalties stacking per agent in coordinated systems. A hearing is set for Oct 5. If you deploy agents that touch NYC, a concrete compliance surface is taking shape.
Claude Code can now audit your CLAUDE.md
Claude Code 2.1.283 adds `/doctor prompt-audit` to review your CLAUDE.md (and AGENTS.md) files for issues, plus `deniedModels` and `availableModelsMatch` settings that pin exactly which model versions your agents may run. MCP tool outputs now surface in OpenTelemetry spans too. Small quality-of-life wins for anyone standardizing agents across a team.
OpenAI open-sources a mental-health eval
OpenAI released MentalHealthBench, an open benchmark of 1,215 conversations built with 80+ clinicians across 22 countries, to grade how models handle sensitive mental-health chats. It's freely available, so if you ship consumer-facing chat you can test your own stack against real clinical rubrics instead of guessing.