Claude Science brings sources, code and review into one workspace

Anthropic has released Claude Science, an AI workspace for scientific research. The service gathers, among other things, literature search, code, figures and connections to research databases in the same environment.

According to Anthropic, each result receives a review history. A figure can be linked to the code and environment that created it, a description of the workflow and the previous conversation. It should make it easier to follow how a result has been produced and to repeat the analysis.

Claude Science can be run locally on macOS and Linux or connected to a remote computer and a compute cluster. Anthropic states that large datasets can stay in the researcher's own environment and that only the context needed for a step is sent to Claude.

That promise does not replace an own check of agreements, authorizations and data flows. A research organization still needs to decide what material may be used, who can connect external resources and how results are to be validated.

The service is in beta for Claude Pro, Max, Team and Enterprise. Team and Enterprise accounts require an administrator to enable it. Anthropic also describes a review agent that can flag incorrect citations and calculations, but an automated check is not the same as independent scientific review.

The saved Instagram signal also claimed that Anthropic would develop its own drugs. It is not supported by Anthropic's product notice and is therefore not included in AI Kollen's reporting.

Därför spelar det roll

When AI helps with research, it's not enough that the end result looks convincing. The source, the calculation, the code and every change needs to be traceable. The same principle is also useful in smaller businesses that use AI for analysis and reports.

Det här kan du göra

1. Check which data leaves your own environment before activating the service.
2. Require figures and conclusions to be traceable to sources, code and calculations.
3. Have a responsible person review the results before they are used in research or decisions.