Claude Science: Anthropic builds an AI workbench that keeps research data on your own machines
Anthropic's new app for researchers bundles 60+ scientific skills, a citation-checking verification agent — and runs locally, so sensitive data never leaves the lab.
Anthropic has released Claude Science, a dedicated AI workspace for researchers — and the most interesting design decision has nothing to do with intelligence. The app runs locally on macOS or Linux and connects to a lab’s own servers and computing clusters, so sensitive research data never has to leave the building.
The workbench pulls together dozens of scientific databases, tools, and software packages in one interface. Researchers can analyze literature, run multi-step data analyses, create charts, and draft manuscripts. More than 60 preconfigured skills — ready-made workflows the AI can execute — cover fields like genomics, proteomics, and cheminformatics. A separate verification agent automatically checks citations and calculations, a direct answer to the most common complaint about AI in research: confidently invented references. Under the hood, it taps Nvidia’s BioNeMo toolkit, which ships with specialized biology models, and researchers can save their own pipelines as reusable skills.
The local-first architecture deserves a closer look, because it solves a problem that has kept many scientists away from AI tools entirely. Unpublished data, patient records, unpatented compounds — none of it can simply be pasted into a chatbot. Claude Science works the other way around: the app operates where the data lives, over SSH or on high-performance computing clusters, and only the specific context Claude actually needs gets sent to the model. When a job needs more power, it scales from a single GPU to hundreds.
What’s behind it: Anthropic is betting that science becomes one of AI’s flagship use cases, and it’s competing for researchers’ trust, not just their subscriptions. The local-first pitch is aimed squarely at institutions whose data policies forbid cloud AI tools. Worth noting the limits before you get excited: the model queries themselves still go to Anthropic’s servers — this is not offline AI, it’s data-stays-home AI. And it’s a beta, so rough edges are guaranteed.
What this means for you: If you work in research — or study anything lab-adjacent — this is worth a look now rather than later: Claude Science is in beta for Pro, Max, Team, and Enterprise users, and Anthropic is backing up to 50 research projects with up to $30,000 in credits each, with applications open through July 15. If you’re not a scientist, the pattern is still the story: “the AI comes to your data, not the other way around” is the design that makes AI usable for medicine, law, and anywhere else privacy actually matters — expect to see much more of it.
Sources
Source: https://www.anthropic.com/news/claude-science-ai-workbench
Using Fable 5 with a Claude subscription? Monday changes how you pay
Fable 5's inclusion in Claude Pro, Max, and Team plans ends July 7. After that, the model is billed through usage credits — here's what to check before Monday.
