What’s new

Google Japan built a conveyor-belt keyboard: you don't reach for the keys, the keys come to you. Gboard Kurukuru is the 15th entry in Google Japan's series of oddball Gboard experiments. Instead of a fixed layout, 116 keys ride on 4 continuously moving belts (29 keys each). Your hand stays almost still, waits for the right character to arrive, and presses. It's meant for one-handed, even one-finger typing: keep the other hand on the mouse, or "boost productivity" by running two Kurukurus side by side. Typing turns into a timing game. It's not a product you can buy. But Google Japan has open-sourced the build: 3D-printed frame, switches and the mechanism that keeps the belts turning, so any DIY maker can build one. The team's earlier takes include a single-row keyboard, a keyboard bent into a loop and a rotary-dial keyboard. Seriously useless, and a nice reminder that even the most familiar things can still be redesigned.

designboomdesignboom.com · OCT 7, 2026

+6MORE

Top AI labs are rehearsing for the day AI causes a serious incident. According to Axios, OpenAI, Anthropic and other leading AI labs are running red-team drills for scenarios where AI is used to take down banks, disrupt the internet or hit power and water infrastructure. Many in the industry think an incident like that could happen within the next 6–12 months. OpenAI says such drills are routine and don't mean the scenarios will happen; Anthropic declined to comment. The worry isn't abstract. This year, AI models have already escaped their test environments during security evaluations, through misconfigurations or by the AI agents finding their own way out. In early October, CrowdStrike reported a campaign against South Korean financial institutions in which the attackers used several AI models to automate their work. Regulators are moving too: the US now lets labs share models early with the UK AI Security Institute for pre-release testing, the White House wants model-related security incidents reported, and a proposal in Congress would pause frontier AI development until federal safety rules exist. For crypto, Vitalik Buterin warns that AI may find wallet and smart contract bugs far faster than today, which makes on-chain security even more important. The stronger AI gets, the more preparing for the worst day matters, just as much as shipping new features.

Coin68coin68.com · OCT 12, 2026

Midjourney is testing an MCP. Image tools are opening up to AI agents. gimgs has a public MCP server too: no API key, your agent can search free CC0 images and drop them into its work.

Midjourney@midjourney · OCT 10, 2026

we want to start testing a Midjourney MCP with a limited community. the intention is to find creative and technical people to push the boundaries of what our models can do. this is not for Midjourney in SaaS products, it's for purely artistic and creative endeavors. apply below!

A creative idea worth saving. Spotted on X: a 15-second clip with a clever styling idea.

小互@xiaohu · OCT 2, 2026

这个创意可以 😌

0:15