Written by Lisa Loud
How Trusted Execution Environments and privacy-preserving infrastructure can unlock the power of medical AI without compromising patient trust.
Artificial intelligence holds the power to knock diseases out one by one. In the landscape of human mortality, roughly one hundred major diseases account for the vast majority of deaths globally. The complexity of these illnesses has historically outpaced human cognitive capacity, but AI—with its unprecedented ability to ingest vast datasets, recognize subtle patterns, and simulate molecular interactions—offers a paradigm shift. We can realistically imagine a world where cancer, heart disease, and neurodegenerative conditions are no longer existential threats.
Yet, this utopian vision is currently blocked by a fundamental paradox: the data required to train these life-saving models is exactly the data we must protect most fiercely.
As the Executive Director of Secret Network, I spend my days working at the intersection of privacy-preserving technology and decentralized systems. What I see in the current healthcare AI landscape is a dangerous reliance on an outdated, centralized data model. If we build healthcare AI on the same foundational architecture that powers today’s ad-driven internet, we risk transforming our most intimate biological realities into commodities.
We are at a critical juncture where AI will either compete with us or collaborate with us. If we allow artificial intelligence to be created with backdoor access for big tech companies or centralized authorities, we are giving away much more than our search histories or purchasing habits—we are giving away our very essence.
The Limits of Centralized Medical AI
Currently, AI models are largely trained on public or siloed datasets. In healthcare, this leads to model saturation and biased outputs. A diagnostic AI trained exclusively on data from a single hospital system or demographic will inevitably fail when applied globally. To build truly effective, unbiased medical AI, we need access to a diverse, global pool of patient data.
However, healthcare organizations are rightfully hesitant to share Protected Health Information (PHI). Traditional data sharing requires moving sensitive information across networks and exposing it during computation. Even with anonymization, the risk of re-identification is high. When AI systems ingest PHI, they often retain traces of that data in their weights and parameters, creating a massive liability.
Furthermore, we must consider the metadata. It is not just the content of a medical consultation that matters; the metadata—who is querying an oncology AI model, how often, and from where—can be just as revealing. In a centralized system, this metadata is inherently exposed to the platform operator.
Confidential Computing: The Technical Imperative
The solution to this paradox is not to halt AI development, but to fundamentally change its infrastructure. This is where Confidential Computing, specifically through Trusted Execution Environments (TEEs), becomes essential.
Confidential Computing ensures that data remains encrypted not just at rest and in transit, but during processing. TEEs act as secure enclaves within a processor. When a hospital sends patient data to an AI model running in a TEE, the data is decrypted only inside this secure hardware environment. The AI can process the data, learn from it, and output the result, but the host system, the cloud provider, and even the AI developer cannot see the raw data.
This technology is foundational to what we are building at Secret Network. By leveraging TEEs, we enable high-performance computations while ensuring complete data privacy. In a healthcare context, this means multiple research institutions can collaboratively train a single AI model on their combined datasets without ever exposing their patients’ underlying records to one another or to the AI creator.
This approach, known as federated learning enhanced by confidential computing, unlocks the vast reserves of private medical data that currently sit dormant. It allows researchers to train models on millions of diverse patient records securely, eliminating bias and dramatically accelerating the pace of medical discovery.
Building a Collaborative Future
If we build AI for competition—prioritizing data hoarding and centralized control—it will ultimately exploit us. But if we build AI for collaboration, embedding privacy from the ground up, we can work alongside it to improve human life.
The transition to confidential AI infrastructure is not merely a technical upgrade; it is an ethical imperative. We cannot afford a future where privacy becomes a luxury good, where wealthier individuals can afford to protect their health data while others are pressured to trade their biological privacy for access to medical AI services.
By demanding privacy-preserving infrastructure, the healthcare industry can lead the charge in responsible AI development. We can create a secure data economy where patients maintain ownership of their information, researchers have the data they need to cure diseases, and AI serves humanity securely and ethically.
The technology to protect our data while unlocking AI’s potential exists today. The only question is whether we have the collective will to implement it before the architecture of our medical future is finalized. Every time a new healthcare AI initiative is announced, we must ask: is this built for extraction, or is it built for collaboration? The answer will define the future of medicine.
