Anthropic previews a Model Hardware Standard letting AI agents run lab and manufacturing equipment

Anthropic has opened a research preview of the Model Hardware Standard (MHS), a shared specification that lets AI agents safely operate physical lab and manufacturing equipment, including microscopes, liquid handlers, and robotic arms. The company is sharing an early version of MHS with a first group of scientific research labs and advanced manufacturers, spanning biotech, robotics, quantum computing, and related fields.

What's new

MHS is a unified driver interface that standardizes how AI agents talk to physical instruments. According to Anthropic, it typically takes a lab or manufacturing facility weeks, if not months, to wire up software so an AI system can control its hardware; MHS is built to cut that down.

"MHS reduces this integration work to hours or minutes" compared with that typical weeks-to-months timeline, per Anthropic's announcement.

Key properties of the standard:

• Device-agnostic. "MHS works with any device that has a programmable interface." It is also model-agnostic, so any agent harness can reach it through standard protocols such as the Model Context Protocol.
• Parallel operation. Agents can run multiple instruments at once, handling tasks from routine drug-discovery experiments to laser calibration on a quantum computer.
• Autonomous recovery. Agents can reason through each step of an experiment, update parameters in real time, and in some cases recover from hardware errors without a human stepping in.

Named research-preview partners include Genentech, the University of Washington, Carnegie Mellon University, HHMI Janelia, QuEra Computing, and Tetsuwan Scientific. The effort began as a collaboration between Anthropic and HHMI Janelia Research Campus.

Context

Anthropic has spent 2026 pushing Claude further into scientific workflows, from the Claude Science product line to its AI for Science program offering free compute credits to researchers. MHS extends that push from software-only research assistance into direct control of physical instruments — a harder problem, since a coding agent that makes a mistake can be rolled back, but an agent driving a robotic arm or a liquid handler cannot.

That's also why Anthropic is framing this explicitly as a research preview rather than a general release, sharing it with partners across science, robotics, electronics, and manufacturing to build safety evaluations and best practices ahead of eventually open-sourcing the standard. The standard is being hardened alongside its rollout rather than after the fact.

Why it matters

Lab automation has long been bottlenecked by bespoke integration work — every instrument vendor has its own control software, and connecting that to an AI system has historically meant custom engineering for each device. If MHS lives up to its billing, it turns that into a one-time integration cost per device rather than a per-lab, per-project cost, which would meaningfully lower the barrier to running AI-orchestrated, round-the-clock experiments.

It also signals where Anthropic sees the next competitive edge for AI agents: not just writing code or answering questions, but directly operating the physical infrastructure of scientific and industrial work. Partners like Genentech and CMU give the preview immediate credibility in biotech and robotics, two fields where the cost of an agent's mistake is measured in broken equipment or ruined experiments — which is exactly why the safety-evaluation framing sits at the center of the announcement rather than as an afterthought.