Sigfox South Africa, a network operator that provides low-power wide-area network connectivity for the internet of things, has applied to join an application-only research preview for a new artificial intelligence specification. The operator seeks to test the Model Hardware Standard released by Anthropic, an artificial intelligence developer that builds agents such as Claude, to bridge physical sensors with AI workflows. Anthropic developed the specification in collaboration with HHMI Janelia Research Campus and introduced it on August 27, 2026, as a shared framework for AI agents to operate physical equipment.
Connecting hardware to digital platforms requires bespoke software integration because machines use varied payload structures and distinct data models. The Model Hardware Standard functions as a driver that allows a device to describe itself in a structured format. This interface dictates the measurement capabilities and operating limits of the hardware. Developers access the model-agnostic framework through the Model Context Protocol alongside standard command-line and application programming interfaces.
“AI has changed the way we interact with the data coming out of environments, both from a historical reporting function as well as predicting potential future events,” says Gregory Rood, CEO of Sigfox South Africa. “This has been made possible by the ability to consume data and generate multiple different outcomes or reports at a rate not humanly possible a few years ago. Therefore the idea of creating a standardised platform that could potentially allow for plug-and-play devices to be integrated into an AI-enabled universe on the go has so many exciting opportunities for business in Africa that we, as the only true open-access IoT network, felt it extremely important to engage in this initiative and provide a blueprint at the outset of what we provide as an LPWAN network operator.”
The global Sigfox ecosystem transports small amounts of device data and connects more than 14 million devices across more than 70 countries. The proposed hardware standard functions as an abstraction layer above existing certification processes and payload structures. Through this setup, a connected object such as a water meter exposes its unit of measurement and operating parameters alongside its primary reading. An artificial intelligence agent could read this context to identify abnormal consumption or potential leaks without custom software translation.
Anthropic limited the initial research preview to scientific research laboratories and advanced manufacturers to establish safety evaluations before making the standard open source. As the specification matures, hardware engineering teams could incorporate machine-readable interfaces at the design stage. This integration would allow environmental monitors and industrial tracking equipment to bypass database storage and supply context to artificial intelligence systems.
