Anthrophic Claude Physics Team Overview and Access Guide
IT Services Physics has created a Claude Team, which is open to any University staff member to join, we just need a cost code.
A Claude Team membership (Standard account) costs £180+VAT per year. Note that our internal billing (and your subscription) will be for the balance of the annual subscription to 12 March 2027, which is our next renewal date with Anthropic.
Request access
Use our Jira Service Management form to request membership of the Physics Claude Team:
https://camjira.atlassian.net/servicedesk/customer/portal/1/group/611/create/2156
About IT Services Physics
The Claude Team is managed by IT Services Physics. We are currently providing IT services to 24 external institutions across the collegiate University, in addition to the Department of Physics, including our signage.phy digital signage system - which is directly deployed at 14 institutions and indirectly at several more. Our grassroots IT service model supports University customers in adopting value-driven IT solutions.
About our Claude Team
Data privacy
Our Data Processing Agreement (DPA) with Anthropic states “Anthropic may not train models on Customer Content from Services". This is contractually binding.
Note that as of 8 June 2026 Anthropic published a new privacy policy (here is the previous version). The broad summary of this is that the scope of the public privacy policy has been narrowed to individual customers (from individual and commercial), and commercial customers are now subject to their own Data Processing Agreements (DPAs) - which may be negotiated on a per-organisation basis.
Please note that Free and purchased Pro Claude Plans do not afford the same privacy protections as our Team plan.
Core capabilities
Chat: High‑quality conversational assistance for literature review, drafting summaries, experimental planning support, documentation editing, and general Q&A. Link
Claude Code: Software development assistance including code generation, debugging, unit‑test scaffolding, translating between languages, and explaining scientific code (e.g. Python/NumPy, MATLAB, C/C++). Link
Claude Cowork: Desktop/agentic jobs can be handed off to Claude to work on autonomously and companion tool. Link
Claude Science: Customisable app (currently macOS and Linux only) that integrates tools and packages scientific researchers use to run analysis and tracing. Link
Claude Design: Create design prototypes, presentation slides (PowerPoint compatible and editable) - exports to Canva, PDF, PPTX, HTML - and can handoff to Claude Code for development. Link
Projects (Team): Structured workspaces to organise prompts, files, notes, and task context per project. Projects allow collaboration among project members to support shared workflows and resources. Link
University collaboration benefits
Shared workspaces and skills: Create project spaces (e.g. “Quantum Materials modelling”, “Detector pipeline”) where members can share prompts, approaches, evaluations, and reusable patterns that improve over time.
Common prompt libraries: Publish departmental prompt sets for tasks like data cleaning, figure annotation, simulation parameter search, or safety/ethics pre‑checks—so groups reuse vetted prompts instead of reinventing them.
Cross‑group teaming: Invite collaborators on specific projects while keeping departmental administration, permissions, and billing centralised.
Onboarding efficiency: New team members can adopt tested workflows (e.g. coding style guidance, analysis notebooks, LaTeX draft aides) to become productive faster.
Team administration and governance
Centralised member management: IT Services Physics manages invitations, removals, and licence assignments.
Best‑practice guardrails: Configured for responsible use, data sensitivity, and reproducibility—supporting safe application of Claude to research and teaching.
How collaboration with Claude accelerates Physics work
Below are practical, discipline‑oriented examples of how teams in Physics use Claude to collaborate effectively.
Research code acceleration: Pair‑program complex analysis pipelines; standardise code templates; generate unit tests and docstrings to enforce quality across projects.
Method sharing: Turn one group’s well‑tuned prompt-and-checklist for e.g. Monte Carlo post‑processing into a reusable pattern for the whole department.
Paper drafting support: Maintain a shared Cowork project with outline prompts, LaTeX snippets, and figure‑caption review prompts that groups reuse across submissions.
Reproducibility: Store the exact conversational context, prompts, and scripts used to derive a result within a project so new members can reproduce and extend analyses.
AI.PHY - Cambridge-hosted for data privacy and digital sovereignty
We are planning to launch AI.PHY in the second half of 2026. It provides local AI first by design, and a cloud AI gateway for those who need one. How this is different from the large cloud AI providers:
Digital and data sovereignty - lets you keep your data in Cambridge, and provides and usable AI for those who can't use cloud AI for compliance reasons.
Privacy - all your chats are encrypted under your own passphrases - we can’t read them. Our AI models are also fixed - there is no training on your data (often a cloud AI "feature") - your chats are not going to be used for someone else's output.
Obfuscation of data to external providers - if you use AI.PHY and choose to route to an external cloud AI, our system removes personal information such as CRSids and names from your input, queries the cloud provider with anonymised tokens substituted for these, and then places these back into the cloud response locally, giving you readable output. This means AI.PHY is better for data and cybersecurity, even if you just use it as a front-end for a cloud provider.
Local ingestion of documents - you can add to our fixed models by adding your own documents - as an individual, in a shared group (you choose), or as public - open within the University. This can provide some local AI expertise with local data we don't want in the cloud on a trillion-dollar company's servers. That might be academic research data, but it could be finance and HR files - for example. Local data connections to Cambridge systems may follow later.
Specialised LLMs - we aim to support specialised LLMs which meet the needs of the University community. We are currently testing a Physics-trained LLM to include in AI.PHY.