← Latest AI News

August 28, 2026 · Daily brief

Anthropic wires Claude into your factory machines

Sovereignty angle
Claude learned to run a laser by trial and error, then wrote its own automation script. Nobody coded it — the machine taught itself. When your equipment takes instructions from a model you rent, who actually owns the factory?

Anthropic launched Model Hardware Standard, letting AI agents control microscopes, robot arms, and lab equipment. Setup time dropped from weeks to hours.

The machines just got a new boss

Anthropic released Model Hardware Standard (MHS) on August 27, a plug that connects Claude to physical machines. Microscopes, robotic arms, assembly lines — anything with a programmable interface.

Setup used to take weeks of custom code. MHS cut it to hours or minutes. Machine owners describe their equipment in plain language. The AI reads it, learns the controls, and operates the device.

In one test, Claude taught itself to align a laser through trial and error, then automated the task in a single pass. Genentech sent Claude a PDF of an experiment; the AI ran it autonomously on real hardware. QuEra Computing reported a 99.3% success rate on laser tasks, up from 58%.

Who's on board

AWS, Hugging Face, Raspberry Pi, and Danaher are testing it. Anthropic plans to open-source MHS after the preview. It's model-agnostic — you're not locked to Claude.

Why it matters

This isn't software automating software anymore. It's AI deciding how your physical equipment moves. If the model changes its rules or raises its price, your factory notices. The machines still belong to you. The brain running them doesn't.