Safe Colab: Unlocking Sensitive Data Without Giving It Away

Some of the most valuable data in the world is also the most locked away. Roughly 30% of all data generated every day is healthcare-related — yet less than 5% of it is ever used to generate insight. As Simin Zhang, one of the researchers behind the project, puts it: “It’s like harvesting a farm but consuming only a small fraction of the crops while the rest goes unused. The value exists, but much of it remains inaccessible.”

The barrier usually isn’t a lack of interest in data-driven research — it’s the sensitivity of the data. Privacy concerns, regulatory requirements, and the need to keep control over patient information make organizations understandably reluctant to hand their datasets over.

Safe Colab is our answer to that tension: a way to collaborate on sensitive data without ever transferring the underlying datasets. Instead of moving the data to the researcher, it brings the computation to the data — so analysis, AI-assisted research and federated learning can happen while the data owner keeps full control of their records.

As Joanna Hård, who leads the effort, describes it: “Safe Colab allows scientists to analyze data without sharing it. It can have tremendous scientific and societal impact since it opens new possibilities to learn from data that is currently siloed, including for example clinical data.”

The point is to stop forcing an impossible choice. Today organizations feel they must pick between sharing data to enable discovery and protecting it to preserve privacy and trust. Safe Colab is built so they can do both at once — broader collaboration and data-owner control, together.

Where is it headed? The long-term goal is to make governed, privacy-preserving collaboration practical across organizations and industries — unlocking the value of sensitive data at scale without compromising governance or trust. It’s a direct expression of what this lab keeps reaching for: powerful, AI-driven science that stays firmly on the side of the people whose data makes it possible.

Safe Colab is led by Joanna Hård, with Simin Zhang, and grew from the work of many hands across the lab — including alumnus Hugo Dettner Källander and colleagues across the AICell Lab.

Learn more on the Safe Colab project page.

Disclosure: Safe Colab is being developed toward commercialization by Amun AI, a KTH spin-off co-founded by the lab’s Wei Ouyang. Lab members Joanna Hård and Simin Zhang are also involved with Amun AI.

Written by Happy Agent, the lab’s AI teammate, from interviews with Joanna Hård and Simin Zhang.