Quick answer
Claude Science is in beta for Pro, Max, Team, and Enterprise users on macOS and Linux. Team and Enterprise users need an administrator to enable it. It can work locally, over SSH, or through an HPC login node, and Anthropic describes supported on-demand compute through Modal.
Each layer requires a separate check: Claude plan and seat, beta availability, operating system, administrator policy, remote authentication, cluster permission, job scheduler, data flow, compute price, retention, and review.
Access checklist
| Layer | Verify |
|---|---|
| Claude account | Plan, seat, region, age or organization requirement, beta state, price, usage limits |
| Organization | Admin enablement, identity, role, SSO, audit, retention, data settings, approved research use |
| Workstation | macOS or Linux version, hardware, disk, packages, secrets, updates, local data classification |
| SSH or HPC | Host identity, user account, MFA, bastion, scheduler, partitions, quotas, modules, storage, logs |
| On-demand compute | Provider account, credential, region, GPU, rate, cap, network, retention, deletion, support |
The Claude Science product guide provides the complete product record.
Team and Enterprise enablement
Administrators should begin with a small approved group and read-only or non-consequential workflows. Define permitted projects, data classes, connectors, skills, packages, remote hosts, and compute accounts.
Review how prompts and the context needed for each step are sent to Claude. Anthropic says large datasets can remain on lab infrastructure, but that statement still requires a concrete architecture and contract review.
Local and remote compute
Use separate credentials with least privilege. Do not give a workbench unrestricted shell, scheduler, storage, network, or cloud rights. Require confirmation before new resources, pin environments, cap jobs and cost, and stop on unexpected host, path, package, data, or output.
For HPC, record scheduler script, modules or container, resource request, input checksums, output paths, job ID, logs, environment, and cancellation. Confirm that the login node is an approved orchestration point and that compute runs on the intended worker resources.
Sensitive data boundary
Classify and minimize data before use. Document what remains local, what context is transmitted, which service receives it, where it is processed, and how long it persists. Protect patient, participant, genomic, proprietary, controlled, export-sensitive, and other regulated information through the lab’s established governance.
Do not interpret reviewer-agent checks or local execution as a compliance certification.
The reproducible workflow guide covers analysis records and independent validation. The Microsoft comparison covers enterprise-platform alternatives.
Frequently asked questions
Which plans include Claude Science?
Anthropic’s launch lists Claude Pro, Max, Team, and Enterprise. Team and Enterprise users need an administrator to enable Claude Science. Verify live plan, seat, region, price, limits, terms, and beta status.
Does Claude Science run on Windows?
Anthropic’s launch lists macOS and Linux, not Windows. Verify the current download page and system requirements rather than inferring support from other Claude products.
Can Claude Science use an HPC cluster?
Anthropic says Claude Science can work over SSH or from an HPC login node and can plan jobs for a lab’s existing compute. The lab still controls authentication, scheduler, quotas, data, software, approvals, monitoring, cost, and safety.
Bottom line
Paid-plan eligibility is only the first gate. Safe use depends on administrator approval, supported systems, least-privilege compute, explicit data flows, versioned environments, budgets, and scientific review.
Official source
Source check: August 2, 2026. Verify the live app, plans, systems, compute, data, pricing, and beta documentation.