Kurt Roloff says large pharma often treats compliance as a cost center, but secure collaboration can turn existing data into a profit enabler. It lets divisions and companies reuse data across geographies, cutting IRB and administrative overhead so rare pediatric cancer research becomes more cost efficient.

Despite processing 15 million clinical charts, Datycs doesn't use this data for model training. Their agreements explicitly respect that data belongs to the patient and the client—an ethical choice that prevents them from building large, aggregated language models from customer data.

Variant Bio's advantage lies in its ethical approach to partnering with indigenous communities. This model, which includes co-designing studies and robust benefit sharing, grants them exclusive access to unique genetic datasets that competitors, focused on traditional data sources, cannot obtain.

Running diagnostic validation studies within regionally concentrated, demographically diverse populations produces clinical data accepted worldwide. Establishing research programs near major diverse population clusters enables startups to bypass the traditional prestige of life sciences hubs like the Golden Triangle.

To lower the activation energy for user adoption, OpenAI deliberately will not use data connected to ChatGPT Health to train its foundation models. This strategic choice is designed to remove any tension between privacy and utility, assuring users their sensitive information is not being used for other purposes and building the trust necessary for scaled impact in the healthcare domain.

A common pitfall is outsourcing patient support (nurses, reimbursement specialists) and inadvertently losing control of the data generated. A critical strategy is ensuring contracts mandate that all data flows back to the pharma company, allowing them to own the patient experience regardless of the service model.

Roloff agrees that AI in medicine is constrained less by models or compute than by access to sensitive data. He says privacy technology is ahead of community acceptance; secure inference through commercial trusted execution environments is emerging, but training large foundational models on patient data remains the 'golden apple' still being pursued.

Sophia Genetics helped a hospital in India go from outsourcing tests to the US (with a 6-week delay) to performing them locally in under two weeks. This approach defines democratization not just as providing access, but as empowering local institutions to build their own knowledge and capabilities.

Actuate Therapeutics demonstrates a patient-first ethos by having a dedicated group that not only helps patients access their own trials but also guides them to other companies' trials if those are a better fit. This builds community trust and prioritizes patient outcomes over proprietary interests.