Anthropic opened Claude Science - a mode where the model conducts several agents across dozens of scientific databases, for real research work. It runs on the lab's own infrastructure, with data privacy by default.
Fact
The facts: access is for paid subscribers, and the program includes grants in compute credits for a limited number of projects, with an application deadline. The idea is specific: not a chatbot for scientists, but a workbench with multi-agent processes over scientific sources.
This one's good, because it's not a promise. A leading lab is shipping AI for science with real grants and data protection - something you can try on a real project this week, not just watch in a slide deck.
Multi-agent processes make sense right here - reading literature, cross-referencing databases, checking hypotheses. And because everything runs on the lab's infrastructure, the data never leaves its place. No compromise here. For once, convenience and privacy aren't fighting each other.
Official primary sources
Anthropic - Claude Science, an AI workbench for scientistsThe visual is generated code-art, no third-party images.