Anthropic Launches MHS Hardware Standard and Claudeforce With Salesforce

MHS is positioned as the hardware analog to Anthropic's Model Context Protocol (MCP). 'What MCP did for software, MHS will do for the hardware world,' said Anthropic's Alek Kemeny. The standard lets Claude understand a device's operating parameters, enabling autonomous experimentation, real-time parameter adjustments, and control over robots and lab or manufacturing equipment without human intervention — a meaningful expansion of Claude's enterprise reach into physical applications.

Anthropic is explicit that this is early: on X the company noted LLMs 'still lack physical intuition, having learned about the physical world from text and images,' and framed the research preview as a way to build safety evaluations and strengthen protections before any open-sourcing of MHS. In parallel, Claudeforce integrates Claude's reasoning with Salesforce's data, workflows, and governance for trusted enterprise action.

The dual move fits Anthropic's enterprise-first strategy at a moment when the company reported a $65B annualized run rate and raised its internal catastrophic-misalignment risk from 'very low' to 'low.' Competitively, it stakes a claim in embodied and lab automation just as Google DeepMind's Co-Scientist begins controlling lab equipment and NVIDIA pushes Cosmos world models for physical AI. The open question is safety: giving a text-trained model direct control of physical hardware raises the stakes of the same agentic-misalignment concerns highlighted by the Hugging Face breach. Watch for the safety-eval results Anthropic promised before wider release.