Claude Science for Healthcare: an R&D workbench, not a clinical tool

What it is

Claude Science is Anthropic's new app for researchers, launched alongside the company's announcement that it will start its own drug discovery program targeting neglected diseases. Claude Science is a customizable app that integrates the tools and packages researchers most often use, produces auditable artifacts, and provides flexible access to computing resources. It runs locally against your own compute, with a reviewer agent that checks the primary agent's outputs — as the pipeline runs, a reviewer agent inspects the outputs, flagging incorrect citations, untraceable numbers, and figures that don't hold up.

If you're a clinician expecting an EHR copilot or a HIPAA-blessed chart summarizer: this isn't that. It's aimed at bench and translational science.

The next-work-session test

Concrete scenario: a translational research pharmacist on a hospital's formulary committee needs to build a rapid evidence brief on a new GLP-1 indication — pulling PubMed, running a small meta-analysis in R, and producing a citations-checked memo before Thursday's P&T meeting.

Claude Science plausibly compresses that from two days to an afternoon, because it runs on your lab's own infrastructure—your laptop, Linux box, or HPC login node—so large or sensitive datasets never have to leave the systems they're already on, and only the context needed for each step of the analysis is sent to Claude. The reviewer agent that flags untraceable numbers is the part that matters for clinical accuracy — you get an artifact you can defend, not a chat log.

What it does not change: anything touching identifiable patient data in Epic or Cerner. That's a different product line.

Pricing

Per Anthropic's plan pages and third-party writeups, Claude Science does not have separate pricing — it ships with existing Claude subscriptions. Claude Science is included in the beta for Claude Pro, Max, Team, and Enterprise users. Pro is listed at $17/month annually or $20/month monthly. Max starts at $100/month. Team standard seats are listed at $20/seat/month annually or $25 monthly, and Team premium seats at $100/seat/month annually. Enterprise pricing is quote-only on the Claude pricing page.

Verified as of publication. Beta status means access and limits can change.

What we'd actually use it for

Narrow it down. Realistic uses for a healthcare professional in R&D or medical affairs:

• Literature triage against PubMed with citation-back-to-source, since PubMed provides access to millions of biomedical research articles and clinical studies.
• Reproducible analyses in R or Jupyter on de-identified trial data sitting on your own HPC node.
• Drafting evidence summaries where the reviewer agent flags fabricated numbers before your name goes on the memo.
• If your org uses Benchling, pulling structured experiment records — Benchling connects Claude to Benchling R&D platform data, enabling scientists to ask questions and receive clear summaries with links back to source experiments, notebooks, and structured records—all while maintaining existing access permissions.

Limits

The big one for clinicians: do not put PHI into it under your Pro or Team seat. Anthropic's own docs are blunt — per public reporting on their BAA, Anthropic's own privacy documentation states clearly that its Business Associate Agreement (BAA) "does not cover Workbench and Console, Claude Free, Pro, Max, or Team plans, and other beta or chat products, features, or integrations". A separate HIPAA-oriented track exists via Claude for healthcare and life sciences, which is where prior-auth and coverage-review use cases live — not inside Claude Science.

Other limits:

• No native EHR integration. No Epic, Cerner, or Meditech connector shipped with Claude Science. Benchling and PubMed, yes; charts, no.
• Beta. Feature set and rate limits will shift.
• Reviewer agent ≠ clinical validation. Flagging an "untraceable number" is not the same as a clinician-in-the-loop sign-off. You still own the memo.
• Local compute is a feature and a tax. Running on your HPC node means your IT team is involved before your researcher is.

Try it if

• You do drug discovery, translational, or medical-affairs R&D work with non-identifiable data.
• Your team already lives in Jupyter/R/Benchling and wants an agent that respects those tools.
• You need auditable outputs for regulatory or committee-facing documents.
• You have infra people who can wire it into your HPC.

Skip it if

• Your job is patient-facing and everything you touch is PHI — wait for a signed BAA on the right SKU, or use the healthcare-specific track.
• You wanted an Epic-integrated scribe or coder. Different product.
• You need production-stable tooling this quarter; beta is beta.
• You don't have local compute to point it at — the "runs on your infrastructure" pitch cuts both ways.

---

This content is for informational purposes only and is not medical advice. AI tools used with patient data must meet your organization's HIPAA and privacy requirements.