When will MHS be available to the public?

Anthropic has started a research preview with select partners. It plans to open-source MHS after completing safety evaluations, but no specific date is given.

What types of devices can MHS control?

MHS can control any device with a programmable interface, such as microscopes, liquid handlers, and robot arms. Devices without such an interface are not yet supported.

How does MHS improve on current methods?

Previously, connecting devices required custom implementations that took weeks to months. MHS uses standard drivers with basic read/write commands and natural language tags, reducing setup to hours or minutes.

Large Language ModelsRoboticsITmedia AI＋Aug 28, 2026Aug 28, 2026, 16:01 JST

Anthropic unveils MHS to let AI control lab hardware

Send on LINE

3 Key Points

What happened
Anthropic announced the research preview of Model Hardware Standard (MHS), a common specification for AI agents to safely operate physical devices, on August 27. It is initially offered to select partners in scientific research and advanced manufacturing. MHS was developed with HHMI Janelia Research Campus and allows agents to control microscopes, liquid handlers, and robot arms in parallel.
Why it matters
Previously, integrating lab and manufacturing equipment took weeks to months because each device had a different programming interface. MHS reduces this integration effort to hours or minutes by using standard drivers with basic read/write commands, making devices discoverable on a network. Early adopters include Genentech, which automated a protein assay with Claude autonomously optimizing liquid handling and recovering from errors, and Carnegie Mellon University, which ran a dose-response test about 3 times faster, finishing in 8 hours versus weeks if outsourced.
What to watch
Anthropic plans to open-source MHS after refining safety evaluations and protections for AI in the physical world. The standard is model-agnostic and works with any programmable device, but currently cannot handle devices without a programming interface. Vendors like AWS, Tecan, Universal Robots, Hugging Face, and Raspberry Pi are adding support.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Get news like this every morning, free

Context & Analysis

Anthropic's MHS addresses a key bottleneck in lab automation: the fragmentation of device interfaces. By standardizing how AI agents interact with physical hardware, it aims to make laboratory and manufacturing processes more efficient. The involvement of major partners like Genentech, Carnegie Mellon, and AWS suggests practical validation beyond a theoretical framework.

The standard relies on MCP, a protocol for AI models to access tools, and exposes device capabilities via auto-generated reference files. Early results show significant speedups, but challenges remain, such as Claude's limited spatial reasoning and the need for human oversight. Anthropic acknowledges these limits and plans to strengthen safety measures before open-sourcing MHS, indicating careful consideration of real-world deployment risks.

FAQ

When will MHS be available to the public?: Anthropic has started a research preview with select partners. It plans to open-source MHS after completing safety evaluations, but no specific date is given.
What types of devices can MHS control?: MHS can control any device with a programmable interface, such as microscopes, liquid handlers, and robot arms. Devices without such an interface are not yet supported.
How does MHS improve on current methods?: Previously, connecting devices required custom implementations that took weeks to months. MHS uses standard drivers with basic read/write commands and natural language tags, reducing setup to hours or minutes.

ITmedia AI＋Read Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.