Anthropic unveils auditable Claude Science AI workbench
Anthropic unveils auditable Claude Science AI workbenchAnthropic has introduced Claude Science, an auditable AI workbench designed to streamline the entire scientific research process, from analyzing. . .
Anthropic unveils auditable Claude Science AI workbenchAnthropic has introduced Claude Science, an auditable AI workbench designed to streamline the entire scientific research process, from analyzing literature to preparing manuscripts for publication. The platform does not represent a new or more powerful AI model, but rather a specialized environment built to run on existing Claude models, including Claude Opus 4. 8, without requiring special access permissions. Claude Science allows users to visualize scientific artifacts alongside the exact code that generated them. The system can display 3D protein structures and chemical molecules, which helps researchers validate and reproduce their work months later. The workbench lets users request edits to figures using plain language—such as removing gridlines or changing an axis to a log scale—and the agent will automatically adjust its own code to comply. Researchers can work with a coordinating agent that has access to over 60 pre-built skills and integrations. These connectors connect to specific databases and software for genomics, single-cell analysis, proteomics, and structural biology. The platform consolidates often-fragmented tools into a single interface, enabling literature reviews, data analysis, coding, and manuscript preparation all in one place. Related:Health plan satisfaction falls as costs climbData privacy is a central design element, with the company stating that information remains on local infrastructure or high-performance computing (HPC) systems. Claude interacts with the connected environment rather than pulling data into the cloud. Users can also save their own workflows as reusable skills to share with colleagues or use for future projects. The platform functions like a research team rather than a single chatbot. As the pipeline runs, a reviewer agent inspects the outputs to flag incorrect citations, untraceable numbers, and figures that do not match the underlying code. It self-corrects as it goes, aiming to reduce errors in the final output. The system runs on a lab’s own hardware, including researchers’ laptops and Linux boxes, and is currently available in beta on macOS and Linux for Pro, Max, Team, and Enterprise plans. Standard AI chatbots often provide answers without showing the steps required to reach them, which can make verifying the work difficult. This gap creates a barrier for scientific discovery where transparency is essential. By displaying the code and environment alongside the final artifacts, Claude Science addresses a specific frustration in the field. The ability to inspect the inputs makes the work easier to validate, ensuring that the scientific record remains intact even as models become more capable of handling complex tasks. Related:Beyond Palermo Sicily’s other side enchantsAnthropic launched Claude Science just weeks after announcing the Claude Fable 5 and Mythos 5 models. These models demonstrated exceptional capabilities in scientific research but required safeguards due to the risks associated with their enhanced power. The company stated that Fable 5 included safeguards where users received responses from the lower-capability Claude Opus 4. 8 for certain topics. Access to these models faced immediate restrictions following their release. On June 12, the Trump administration applied export controls citing national security concerns, forcing Anthropic to disable Fable 5 and Mythos 5 for all users. The government required Anthropic to restrict access to foreign nationals, both inside and outside of the U. S. Anthropic suspended access to the models for all users.