Model Hardware Standard: AI operating physical equipment
- URL: https://www.youtube.com/watch?v=UxJZrCFzTHY
- Channel: Anthropic
- Fetched: 2026-10-19T07:31:09+00:00
- Language: English (auto-generated)
Transcript
[00:00] Now, experiments that took weeks take
[00:03] days.
[00:04] That can enable scientific breakthroughs
[00:06] [music]
[00:07] much quicker.
[00:09] >> For the past year, we’ve been putting AI
[00:11] to work inside labs and manufacturing
[00:14] facilities.
[00:16] We kept hitting the same wall.
[00:19] There’s no common way to connect a model
[00:21] to physical equipment. The Model
[00:23] Hardware Standard [music]
[00:24] changes that.
[00:25] >> MHS gives any AI model one standard way
[00:28] to connect with and operate devices. If
[00:31] you wanted your instruments [music] to
[00:32] work together, say a camera and a stage
[00:34] of a microscope, you had to build a
[00:36] custom software integration between
[00:38] them.
[00:39] Every device you add complicates the
[00:41] picture.
[00:42] For complicated experiments, that’s
[00:44] weeks of work.
[00:46] With MHS, each device connects once
[00:49] through one interface.
[00:50] >> [music]
[00:50] >> The setup is accelerated by the agent.
[00:53] Any device that speaks MHS can talk to
[00:56] anything else that does. Devices
[00:58] communicate at bare metal speed.
[01:01] The agent has access to this context and
[01:03] can control operations.
[01:06] >> Here we’re seeing Claude controlling a
[01:07] Leica microscope on its own. It’s
[01:10] focusing the microscope, looking for a
[01:12] bacteria, and deciding what to capture
[01:14] next.
[01:16] At Genentech, a scientist designed an
[01:18] experiment in a PDF, dropped that PDF
[01:20] into Claude, then Claude autonomously
[01:23] executed that experiment. [music] When
[01:24] things went wrong, it was able to
[01:26] recover even overnight.
[01:28] >> The iteration cycle is much faster and
[01:31] the focus for me as a scientist shifts
[01:34] to actual scientific questions. Once
[01:36] everything is connected, running an
[01:38] experiment looks like you ask Claude to
[01:40] start [music] it.
[01:41] Here we’re looking at images of neurons
[01:45] in the brain.
[01:46] With MHS, I can instruct [music] the
[01:48] microscope to move around, go deeper, or
[01:52] even take side views, all in real time.
[01:55] >> If you can test [music] hypotheses
[01:57] faster, you could create general
[01:59] technologies faster, [music] like
[02:01] finding new materials, also solving
[02:03] important scientific problems. This is
[02:04] how a century of progress can compress
[02:06] into a decade.