Anthropic pushes AI out of chat and into controlled lab equipment

August 28, 2026

Anthropic opened a research preview of its Model Hardware Standard to select scientific labs and advanced manufacturers. The standard is a shared specification for letting AI agents control lab and factory devices with defined safety rules, and Anthropic says it plans to make it open source later. That matters because a common control layer would let equipment makers, labs and model providers build against the same interface instead of each wiring AI to machines on their own.

The move appears to shift generative AI another step from text work into physical operations. If labs and manufacturers can test one shared way to connect models to instruments, the practical question moves from whether an AI can answer well to whether it can act inside governed limits on real equipment. Microsoft shows the same pressure from the other side: once AI reaches business systems and physical processes, the value is less in a chatbot window than in the control surface around it.

Anthropic also expanded its support for scientists with 10,000 seats on a new Claude team plan and broader AI for Science credits. That likely helps seed the users who might later demand the hardware standard, because access to models and access to instruments now start to reinforce each other.

Microsoft made a new Azure DevOps plugin available in the GitHub Copilot app. Microsoft also made its Copilot in 30 trial generally available for cloud partners. Google launched Expert Intelligence in Gemini Notebook. Cohere released Parse, a document-to-Markdown model available through Microsoft Foundry, AWS SageMaker and Cohere's Model Vault. More than 100 technology companies signed an open letter on AI-enabled cyber attacks. LibreOffice 26.8 shipped without generative AI features.