The Aridhia AI Sandbox
The Aridhia DRE provides a controlled environment where researchers can experiment with large language models on sensitive healthcare data, without any prompts, context, or outputs leaving the secure boundary.
Learn MoreUsing AI with sensitive data
Modern AI tools can help with clinical records, genomic datasets, and trial data analysis. But the data itself is subject to strict governance.
The governance gap
Patient confidentiality, data-sharing agreements, and regulatory controls all require that sensitive information stays within controlled environments.
Most AI services work the other way around. When you use a cloud-based AI assistant, your prompts and data leave your environment, travel across networks, and are processed on infrastructure you don’t control.
For general productivity tasks, that’s often acceptable. For healthcare research, it breaks the trust model that makes access to sensitive data possible in the first place.
External AI Services
AI Sandbox
Learn more about the Aridhia approach to responsible and secure use of generative AI in a Trusted Research Environment
A contained space for AI experimentation
The term “sandbox” comes from software development, where sandboxes let developers test code safely before releasing it into production systems.
Isolated environment
A contained space where researchers can run AI models, experiment with prompts, and analyse datasets without affecting anything outside the boundary.
Data never leaves
Prompts, context, and outputs all stay within the secure boundary. The models run on infrastructure inside your environment.
Complete audit trails
Every interaction is logged. Which model was used, what prompts were submitted, what responses came back. The chain of custody is captured.
Reproducible results
Administrators control which model versions are deployed. Results can be reproduced and methodologies defended when required.
Why AI sandboxes matter now
Both the EU and UK are establishing frameworks that emphasise controlled experimentation under supervision. AI sandboxes are becoming a regulatory expectation.
EU AI Act
The EU AI Act requires all Member States to establish at least one AI regulatory sandbox by August 2026. These are structured environments for testing AI systems under regulatory oversight before market deployment.
Healthcare AI systems are typically classified as high-risk, requiring stringent compliance with transparency, safety, and data governance rules. The sandbox framework gives developers a path to demonstrate compliance while testing on real-world clinical data.
Deadline: August 2026
MHRA AI Airlock
The MHRA’s AI Airlock programme is a dedicated regulatory sandbox for AI as a Medical Device. Now in its second phase running through to March 2026, the programme brings together manufacturers, regulators, and clinical partners.
The stated goal is for the NHS to become the most AI-enabled healthcare system in the world, with rigorous testing and evidence as the foundation.
Phase 2: Through March 2026
A complete AI sandbox inside the Aridhia DRE
AIRA is an AI Research Assistant framework embedded within the Aridhia Digital Research Environment, providing offline large language model inference entirely within your secure workspace.
Offline inference by default
AIRA runs a proprietary scheduling framework on workspace infrastructure. Prompts never leave your dedicated secure cloud environment.
Governed API passthrough
External access is off by default. Administrators decide if and when it’s permitted, with every call logged including prompts and responses.
Full audit trails
Every inference job is logged in the DRE’s audit system. The chain from prompt to response is captured for regulatory review.
Administrator control
Data owners decide which models are available, who can access them, and how they’re configured. Model versions are controlled for reproducibility.
OpenAI-compatible APIs
Point existing tools at AIRA’s endpoints. Researchers and developers work with familiar interfaces that run locally instead of calling external services.
Data residency guaranteed
All inference occurs inside the DRE’s locked-down Azure environment. Data residency is demonstrable to auditors and ethics committees.