Anthropic introduced a new interoperable standard called MHS to enable AI agents to operate microscopes, manipulators, lasers and other hardware.

In internal tests, the Claude model adjusted laser settings, analyzed camera feedback and converted the successful procedure into an executable script.

What MHS defines

MHS defines a protocol for device control, telemetry and safety checks so models can interact with diverse equipment across different vendors.

Anthropic claims tasks that previously required weeks or months of integration and calibration could be reduced to hours or minutes with MHS.

Testing with Claude

During demonstration runs Claude autonomously tuned a laser, evaluated images from a camera, iteratively adjusted parameters and finalized a reproducible script.

The company says the same standard is intended to be model-agnostic, allowing non-Anthropic systems to adopt MHS for hardware control.

Safety and oversight

Anthropic emphasizes safety features and supervisory controls in MHS but provides limited technical detail about the guardrails implemented during demonstrations.

Adopters will need organizational policies, physical safety practices and verification processes in place before deploying autonomous agents under MHS.

Availability and implications

Anthropic says the standard can be used by other models, but did not publish timelines or public repositories with the announcement.

Broader adoption could streamline laboratory automation and robotics workflows while raising questions about certification, auditability and operational oversight for deployed systems.