Model Hardware Standard Won’t Drive Autonomous Manufacturing Anytime Soon

Anthropic unveiled its latest Artificial Intelligence (AI) standard for interoperability, Model Hardware Standard (MHS), in August, broadening its research preview scope. Designed to enable agents to communicate with and understand physical machines, MHS provides a new foundation for Physical AI beyond robotics. Designed to turn any device into a self-learning machine, Anthropic is emphasizing manufacturing applications to drive autonomous operations. MHS gives agents the ability to operate on hardware but concerns about deterministic outcomes will reduce utilization in automation tasks. MHS will first serve as a powerful communication protocol to build more interoperable systems, becoming an intelligent Open Platform Communications Unified Architecture (OPC UA) before transitioning into a trusted Physical AI solution.

While it is becoming increasingly difficult to find an industrial automation supplier not interested in embedding AI within an Industrial Personal Computer (IPC) or Programmable Logic Controller (PLC), that has taken the form of building leaner models and stronger hardware to perform tasks disconnected from the machine.  These models may be used for advanced analytics capabilities at the edge, reducing the data transfers that sacrifice context, without autonomously altering control code. Recently, Emerson Electric has taken open-source models that fit on a Graphics Processing Unit (GPU) already supported by its IPC platform to develop PLC code agentically in an air-gapped manner.

MHS-enabled autonomous controllers will struggle even if they manage to achieve deterministic output through the correct harnesses. Human-in-the-loop systems will become the baseline of agentic engineering tasks in automation as adoption grows, because trust will lag behind. Requiring verification of changes that an MHS-enabled agent would make means that there is little to distinguish it from agentic tools that live in the development environment on the market.

MHS will be strongest in the industrial automation communication layer, augmenting and deploying alongside existing protocols OPC UA and Message Queuing Telemetry Transport (MQTT). While these two drive interoperability, integrating multi-vendor systems has been a costly and complex problem. MHS will thrive by reducing integration costs: a self-aware machine will inform how it coordinates with other MHS-enabled machines and drives communication between them but will not change machine state. Here, the integration burden shifts from the integrator to the machine. Industrial automation competitors that emphasize openness, such as Schneider Electric, Bosch Rexroth, and Phoenix Contact, should consider joining the research preview for AI-powered interoperability. Still, vendors with a closed stack have little to gain.

Anthropic is involved in the growing industrial cloud infrastructure layer, with Amazon Web Services (AWS) being a part of the research preview, and it has partners across the industrial ecosystem. Among them, Siemens stands out as a potential industrial automation partner to better develop the technology. Having already expressed an interest in goal-based automation, Siemens is willing to entertain automation systems that are not pre-configured and deterministic. The development of non-deterministic automation would violate how the market currently understands itself and would require an ongoing education effort and trust-building to drive adoption. At the same time, Siemens has already emphasized industrial AI and would be a critical vote of confidence for MHS in industrial applications, providing meaningful access to pilot projects. Other industrial automation competitors with an emphasis on openness, such as Schneider Electric, Bosch Rexroth, and Phoenix Contact, should consider joining the research preview in the context of AI-powered interoperability, but vendors with a closed stack and that maintain their deterministic emphasis should pass on joining the research review.

Process specialists are another avenue for Anthropic to pursue, as Physical AI in the form of intelligent valves and actuators has been a domain of high interest (see ABI Insight “ABB to Acquire Rotork for US$5.5 Billion to Shore Up Intelligent Valve Offering” for more information). Companies such as ABB, Valmet, and Honeywell Technologies are all pushing forward autonomy in process manufacturing.