Anthropic launches Claude Science to turn researchers into lazy project managers

The newly launched Claude Science platform operates on top of existing Claude models rather than introducing a brand-new neural network. It functions as a digital laboratory workbench designed to coordinate multiple AI agents under one master project manager. This head assistant connects directly to over 60 scientific databases, translating natural language prompts into complex computational tasks like protein structure analysis and genomics.

The system is built to delegate work by automatically spawning sub-agents for specific chemical or biological calculations, while a dedicated AI fact-checker cross-references citations and math to prevent the hallucinations that usually plague scientific publications. Crucially, this fact-checker is still powered by the same underlying Claude model, meaning it is essentially grading its own homework.

For researchers worried about intellectual property, the entire workspace can be run locally on lab-owned servers. When generating visual data like 3D protein models, the platform bundles the raw code and exact software environment directly into the image file, allowing other scientists to recreate the exact graphic with a simple text command.

This launch puts Anthropic in direct competition with OpenAI, which recently deployed its biology-focused GPT-Rosalind model to a highly restricted list of corporate partners like Moderna and Amgen. Meanwhile, Google DeepMind dominates the absolute foundation of the field by integrating proprietary models like AlphaFold into its own Gemini for Science suite.

The academic publishing world is about to be flooded with peer-reviewed papers generated, formatted, and verified entirely by machines. If one AI writes the paper, another AI peer-reviews it, and a third AI summarizes it for the public, human scientists might soon find themselves relegated to just securing the grant funding.

Source: Anthropic