Anthropic Opens Powerful Biology AI To Verified Scientists

Anthropic is opening a more powerful version of Claude to verified scientists and life-sciences organisations, attempting to give legitimate researchers more freedom without removing the safeguards designed to prevent dangerous biological misuse.

The company calls it the Life Sciences Verification Program. It is launching in beta for teams and institutions that can prove their credentials, explain their intended work and pass security and ethics checks. Approved users can access Claude Mythos, Opus and Sonnet with more permissive biology safeguards than ordinary accounts receive.

The idea is straightforward. AI models can help scientists review papers, analyse experimental results, design workflows and identify possible drug candidates. Yet the same underlying knowledge can be repurposed to assist harmful biological work. Anthropic’s answer is to separate verified research access from the general product instead of imposing the same limits on everyone.

There will be two routes. Standard Use grants are intended for established research and development, while High-risk Use grants cover work that requires deeper review because of its methods, materials or potential consequences. Access will initially be available through Claude Enterprise, Team and the API, not individual consumer plans.

This is not simply a product launch. It is an early attempt to decide who should receive the most capable scientific assistance and under what conditions. The verification process could enable useful work that broad safety filters currently interrupt, but it also gives a private AI company considerable influence over which institutions qualify.

Anthropic says customer data will be compartmentalised and handled under its enterprise controls. The first version will not support protected health information or provide a BAA-enabled service, so hospitals and researchers working with identifiable patient data will still face important limits.

The programme also arrives as Anthropic develops stronger models for sensitive domains. Its planned Mythos-class systems are being positioned for demanding security and scientific work, where ordinary chatbot rules may be too restrictive but uncontrolled access would be irresponsible.

Researchers will want clarity on how applications are judged, how quickly appeals are handled and whether smaller laboratories outside the United States can qualify on equal terms. A verification system that favours large pharmaceutical companies could widen the scientific divide even while the technology itself becomes more capable.

Still, the programme recognises a real problem. The safest model for the public may not be the most useful model for a trained biologist. Anthropic is trying to create a middle layer between unrestricted access and blanket refusal. Whether that layer earns trust will depend on transparent decisions, strong monitoring and evidence that useful science is not being quietly locked behind institutional privilege.