Anthropic has launched Claude Science, a new AI workbench designed to bring the tools, data sources and computing resources used by scientific researchers into a single environment.

Available in beta for Claude Pro, Max, Team and Enterprise users, the app is intended to support scientific work across areas including genomics, single-cell analysis, proteomics, structural biology and cheminformatics.

Claude Science integrates commonly used research packages and databases, produces auditable outputs and allows scientists to access computing infrastructure ranging from local machines to high-performance computing clusters and on-demand GPUs.

The launch represents Anthropic’s largest expansion into life sciences since the company began focusing on the sector last autumn.

Scientific research frequently requires users to move between separate databases, file formats and specialist applications. Anthropic said Claude Science consolidates these systems into a single research environment capable of supporting literature analysis, multi-stage investigations and the production of figures and manuscripts.

The app can be accessed locally on macOS or Linux, as well as remotely through SSH or an HPC login node.

Users work with a generalist coordinating agent that has access to more than 60 curated scientific skills and connectors. The agent can also create additional agents or work alongside specialist agents developed by researchers.

A separate reviewer agent examines citations, calculations and generated figures, flagging errors and attempting to correct them during the research process.

Claude Science can produce visual scientific outputs including 3D protein structures, genome browser tracks and chemical structures. Researchers can annotate figures and manuscripts directly and ask the system to make changes using natural-language instructions.

For example, a user could request that gridlines be removed from a graph or that an axis be converted to a logarithmic scale. The system then updates the underlying code used to generate the figure.

Each generated output includes the code and software environment that produced it, a plain-language explanation of the process and the message history behind the result. Anthropic said this is intended to make outputs easier to inspect, validate and reproduce at a later date.

The app can also manage computing jobs on researchers’ existing infrastructure. Claude Science drafts an analysis plan and asks for permission before accessing additional resources, allowing the user to review or revoke decisions before a job is submitted.

It can work with a laboratory’s existing HPC cluster through SSH or use a researcher’s Modal account to access on-demand computing resources. Anthropic said analyses can be scaled from a single GPU to hundreds of GPUs when required.

Because the agents operate within an active session that retains context, large datasets may only need to be loaded once. Claude Science runs on infrastructure already used by the laboratory, meaning large or sensitive datasets can remain on local systems, Linux machines or HPC login nodes.

Only the context required for each stage of an analysis is sent to Claude, according to Anthropic.

Researchers can also fork a session to compare different analytical approaches without losing the original workflow.

Claude Science comes preconfigured to access scientific sources including UniProt, PDB, Ensembl, Reactome, ClinVar, ChEMBL and GEO, alongside journals, preprint servers and domain-specific models.

The app also uses capabilities from NVIDIA’s BioNeMo Agent Toolkit to connect with life sciences models and libraries including Evo 2, Boltz-2 and OpenFold3.

Researchers can connect Claude Science to their own datasets, models and validated pipelines. Workflows can be saved as reusable skills, while preferred laboratory tools can be incorporated through connectors and made available automatically in later sessions.

Anthropic said researchers participating in the beta have used the workbench for tasks including single-cell RNA sequencing, CRISPR screen design, protein structure prediction and cheminformatics.

The workflow uses sub-agents to examine thousands of research papers, extract central claims and quantitative findings, and store the information in an evidence database. Other agents then construct the narrative and write individual sections, while dedicated systems generate cross-study figures from the collected evidence.

The process also uses actor-critic pairs, with one agent producing content and another reviewing it for accuracy and citation fidelity.

The beta is available on macOS and Linux. Team and Enterprise customers must have the app enabled by an administrator.

Anthropic has also introduced discounted Claude Team seats for active scientific laboratories at academic institutions and nonprofit research organisations.

The company will support up to 50 Claude Science AI for Science projects with as much as $30,000 in credits for each selected initiative. Modal will provide up to $2,000 in computing resources for some projects.

Anthropic said it is looking for proposals covering a range of scientific fields, with an initial emphasis on biology and biomedical research. Applications will remain open until July 15, with successful projects due to be announced by July 31.

The projects are expected to run from September 1 until December 1, 2026.

Big Week for Anthropic as Fable 5 Returns Following Export Control Reversal

The launch comes as Anthropic prepares to restore global access to Claude Fable 5 after the US government lifted export controls covering the model and Claude Mythos 5.

The restrictions had required Anthropic to prevent foreign nationals from accessing the models, regardless of whether they were located inside or outside the US. Because the order took immediate effect, Anthropic suspended both models for all users.

Over the past two weeks, we have worked closely with Anthropic to analyze and approve Fable 5 to ensure alignment across the US Government and strengthen America’s leadership in AI. https://t.co/PtVJWyQ9kH

— Howard Lutnick (@howardlutnick) June 30, 2026

Fable 5 is due to return globally on July 1 through Claude Platform, Claude.ai, Claude Code and Claude Cowork.

Anthropic said Pro, Max, Team and selected Enterprise users would receive Fable 5 within as much as 50% of their weekly usage allowances until July 7. Access will move to usage credits after that date.

Recommended reading

• Anthropic Investigates Claimed Unauthorised Access to Mythos
• Anthropic’s Mythos Lands It a White House Meeting
• Is Mythos as Dangerous as Anthropic Claims?

The company also intends to restore the model through AWS, Google Cloud and Microsoft Foundry as quickly as possible.

Mythos 5 has not yet received the same broad restoration. Anthropic said access had been returned to a group of US organisations following government approval on June 26, but the company is continuing to work with officials to expand availability to additional domestic and international partners participating in its Glasswing programme.

Anthropic said the restrictions were introduced after Amazon researchers identified a technique capable of bypassing Fable 5’s safeguards and prompting the model to identify software vulnerabilities.

In one case, the model generated code demonstrating how a vulnerability could be exploited. Anthropic noted that other AI models, including Claude Opus 4.8, GPT-5.5 and Kimi K2.7, could identify the same vulnerabilities.

The company has since trained a new safety classifier targeting the behaviour described in Amazon’s findings. Anthropic claims that the specific bypass is now blocked in more than 99% of cases.

Requests blocked by the classifier will be redirected to Claude Opus 4.8.