Anthropic’s MHS is an exciting new standard for instrument manufacturers to aspire to. Crucially, it underscores a growing demand for an integrated compliance and data governance framework — and a solution approach that combines deterministic AI with workflow optimization.
Anthropic’s recently announced Model Hardware Standard (MHS) is a fascinating step in connecting intelligent agents with physical laboratory equipment. By harmonizing how systems discover and interact with instruments, it opens real possibilities for connectivity, standardization and, indeed, repeatability across the bench.
At the same time, bringing AI interfaces — traditionally built on large language models — to physical equipment introduces distinct regulatory challenges, particularly for CROs, CDMOs and hardware OEMs operating under strict oversight. This calls for a compliance layer to govern instrument operations.
At Monopoint Ventures, we approached this challenge from an architectural perspective. Our patent-pending platform, Virgil, is engineered as a Level 3 governance layer first, respecting the ISA-95 standard for enterprise system integration. Level 3 systems record authorship and approval of the procedure, maintain version control over what is effective, and hold a change record that survives an inspection. Virgil enforces built-in validation and produces deterministic, repeatable procedure content, while providing the full power of a conversational interface for authoring the workflow context.
MHS, as currently described, sits naturally at Level 2, helping schedulers and executors do the work on the bench. We think Anthropic’s initial research preview is critical to begin answering the question of how best to reduce friction and improve interoperability between instruments in the laboratory. We also believe Virgil is the missing Level 3 compliance piece for when MHS-enabled instruments go live. How the practical boundary is drawn between the two levels at your laboratory will determine data responsibility and legal liability.
If you are an automation vendor, OEM, translational biotech, pharmaceutical company or contract testing organization, the Level 3 question is going to arrive regardless of which standards you adopt. A procedure spans instruments from several manufacturers, so the layer that governs it cannot belong to any one of them. We have built that layer, and it is running today.
With all the excitement and uncertainty about the future of AI in laboratory operations, please get in touch with questions or concerns about your own situation.