On 27 August Anthropic opened a research preview of the Model Hardware Standard - a shared driver through which an agent discovers the instruments in a lab and gives them commands. Integration that, by the company's account, takes weeks drops to hours. The open-source release comes later, after safety evaluations.
- MHS reduces any instrument with a programmable interface to a handful of simple read and write commands.
- The driver also carries knowledge that used to live in manuals and in people's heads: what the device measures, what can be adjusted, which limits must not be crossed.
- For now it is a closed preview for selected labs and manufacturers; Anthropic plans to open the code after joint safety evaluations.
Many labs have one person who knows how to get the microscope, the camera and the liquid-handling robot to pass data to each other. That knowledge is often written down nowhere. It lives in their head and in a few scripts with names like final_v3.
That knowledge, scattered across heads and folders, is exactly what Anthropic is trying to put into a standard.
What convinces more than the partner list
One detail from their own write-up. Claude moves a laser, watches through a camera where the beam went, and repeats until it understands how the system behaves. Then it packages what it learned into a script that aligns the laser in a single command, without reasoning at every step. That is how a good technician works: try, watch, finally automate it and stop spending attention on it.
The numbers are the partners', from real instruments, not a demo. At Carnegie Mellon, serial dilutions run about three times faster. At QuEra the controller the agent wrote brings the lasers back to the required frequency in 99.3 per cent of cases without human intervention.
Anthropic states the limit itself, and I like that. The model learns the physical world from text and images. At Genentech people had to explain to it that the errors came from foam in the samples, not from a software bug. You do not fix foam with code.
Signing up for the preview makes sense if you work with instruments that have a programmable interface but each speaks its own dialect; an instrument without such an interface stays outside the standard for now.