A free access card for browser, inbox, files and the actions that need you.

A free three-proof handoff for changed files, real tests and reviewable evidence.

A free local search test for docs, code and media—without sending them away.

A decision API, a proof-checking lesson and a free prompt for clean handoffs.

A free, static first pass for authorized Electron and JavaScript artifacts.

Skills replace Gems—and today’s build makes the idea portable, inspectable and free.

Google’s new frontier model is launching to defenders first. The useful reason why.

A safe preflight for the configs, tools and skills your agent can now touch.

Plus: an outcome-first prompt, three free ops tools and browser checkout with a buyer gate.

Plus: Gemini buys on Flipkart, a new scam checker and 3 free building blocks.

Plus: a free prompt to map any messy task into a safe decision flow.

Jev: fastest adoption in Vercel history

Plus: Claude now builds 26% of Anthropic itself

One model told its future self to ignore the rules. Here's the audit I run now.

The new US advisory tells labs to degrade, not block. Here's how to check yours. →

Anthropic overturned its own verdict. California signed. Agents kept leaving.

Plus: the 5-minute log that catches an agent going off-script →

Plus: the Hub vs your own disk, tested for real, and 3 AI jobs →

PLUS: Anthropic files for a ~$2T IPO, DeepSeek hits $74B, Apple names a new CEO

PLUS: Hollywood writers call it "digging the grave," young workers fired 16% more

PLUS: Harness beats model switching by 3x, Z.ai rivals Claude in coding

PLUS: Meta's Muse Glimmer is free, offline, and open source

PLUS: Kimi K3 has no cyber guardrails, and Meta's AI also hacked a company

PLUS: It's 100x cheaper than Claude, and 5 Chinese models dropped in 8 weeks

PLUS: OpenAI's rogue AI hit a second company, Microsoft committed $329B to data centers