What Is Confidential Computing for Healthcare?

While the rest of the business world seems to dive headfirst into the realm of agentic AI, healthcare organizations have to be more cautious. For AI models and agents to be valuable, you have to give them access to your data. But healthcare data is highly sensitive. HIPAA (The Health Insurance Portability and Accountability Act) is no joke — you can’t let any system use your data unless you’re fully confident that it’s compliant and private.

Confidential Computing can help provide that peace of mind and as a result, open the door to new opportunities and projects that may have previously felt too risky.

The Three States of Data

Understanding the importance of Confidential Computing first requires recognizing the three states of data, as it relates to encryption:

• At rest: Anytime your data is being stored and not in use
• In transit: When your data is in the process of being transferred between two points
• In use: Whenever your data is being processed or analyzed by a program or application

Traditionally, the third state (in use) is when your data is most at risk, since data has generally had to be decrypted for a program to access it. In-use data has historically been a major vulnerability gap for regulated sectors like healthcare. But you can’t just not ever use your data. That’s wasting one of your most valuable resources.

That’s the problem Confidential Computing addresses.

What Is Confidential Computing?

Confidential Computing is the practice of encrypting your data while it’s in use — not just when it’s in transit or at rest. Data in-use protection works by creating a Trusted Execution Environment (TEE). A TEE is part of a processor that’s been segregated from the rest and secured with encryption, keeping any data on it confidential, even while analytics or AI programs are being run on it.

Confidential Computing enables you to achieve three core pillars of data security:

• Confidentiality: Any sensitive data used on your TEE is only accessible to authorized parties.
  

• Integrity: In addition to being unable to access your data, unauthorized parties will also be incapable of modifying it. You can ensure that only authorized code runs in the TEE, so any changes made to the data will be intentional and approved by someone with the proper authority.
  

• Cryptographic verifiability (or attestation): You don’t just have to trust that confidentiality and integrity are in place, Confidential Computing also prioritizes verifiability, meaning you can cryptographically prove to others (auditors, clients, security officers in your organization, etc.) how secure the TEE is and that the codes or programs run on it are only those that have been previously vetted and approved.

Threats Mitigated by Confidential Computing

Confidential Computing can provide a level of AI security in healthcare that addresses the main threats you’re worried about. Using TEEs to supply in-use data protection can shield you from:

• External bad actors: The most obvious threat to be aware of is outside hackers who try to breach your internal systems. Hopefully, your system is already designed with protections meant to keep external threats out of your server. But with Confidential Computing, even if a hacker manages to access your larger system, the data within your TEEs will remain secure.
  
• Malicious cloud admins: You’d like to think that outsiders are the real threat and you can trust your employees, but the truth is more complicated. The Ponemon Institute and DTEX’s annual report on insider threats found that malicious insiders caused 6.3 incidents on average in 2026, at a typical cost of $742,125. Confidential Computing gives you the capability to limit which internal users have access to sensitive data, so you can ensure only the people and programs that absolutely need to see that data are able to.
• Neighbor tenant risks: Healthcare organizations that share their cloud architecture with other organizations as a money-saving measure introduce additional risks in the process. Confidential Computing helps reduce the risk, at least when it comes to the data used on TEEs, by placing clear limits on who can access the environment and its contents.
• Host system bugs: Bugs are basically inevitable with technology, and there’s always the risk that your host system will have underlying vulnerabilities that are hard to foresee. Confidential Computing helps limit the risk of these bugs as well by keeping your most sensitive data segmented away from the rest of the system.

Google Cloud’s Portfolio: The Hospital Metaphor

Google Cloud’s Confidential Computing profile offers different levels of secure TEEs for different use cases. Any of these offer enough protection to make HIPAA-compliant cloud AI projects possible, but each is good for different use cases. To understand how each level works, it helps to compare the different options to different parts of a hospital.

Confidential VM, “The Secure Wing”

Google Cloud’s most basic offering for Confidential Computing, Confidential VM (virtual machine) is a single, isolated VM that’s encrypted in memory. It isolates workloads in a hardware-encrypted environment, so you can run applications while keeping your data well protected.

In terms of our hospital metaphor, this functions like a secure hospital wing, set apart from the rest of the building, with limited access to keep patients private and secure. Confidential VM is a good option for lift-and-shift projects and for dedicated, single-tenant workloads like sensitive batch processing and single database servers.

Confidential GKE, “The Hospital Building”

Google Cloud’s GKE (Google Kubernetes Engine) is the next step up from Confidential VMs. It uses the same level of protection, but extends it from one VM to a full Kubernetes cluster.

Where the Confidential VM is one wing, the Confidential GKE is like the main hospital building — it contains multiple rooms and services, but provides a high level of data in-use protection to all of them. Confidential GKEs are a smart choice for microservice-based applications, streaming pipelines (like real-time audio buffering and speech-to-text services), and containerized AI or ML (machine learning) platforms.

Confidential Space, "The Vault Room"

Google Cloud’s most advanced option, Confidential Space, builds on Confidential GKE architecture, but adds features that make it more valuable for multi-party trust. For projects that require collaboration, Confidential Space allows you to run programs on your secure data (and that of your collaborators) without revealing your intellectual property to your partners or the cloud operator.

Confidential Space creates neutral “clean rooms” you can trust, with provable hardware-encrypted AI pipelines. The data in a clean room is only visible to the data owners and any programs that need to access it to complete the necessary work, which means your proprietary AI models stay hidden from outside eyes. In our hospital metaphor, this is like a sealed vault for sensitive records that has extra security layers on top of those active in the rest of the hospital.

Confidential AI in Healthcare

Google Cloud Confidential Computing opens the door to more secure uses of AI. For healthcare organizations, that’s a game changer. Having a valid option for confidential AI in healthcare introduces a number of valuable use cases, such as:

• Ambient clinical scribing: Creating a record of important takeaways after each healthcare appointment is an important step in providing comprehensive care, but it’s a task that adds a lot of work for providers. Using ambient clinical scribing can take this tedious task off overworked physicians’ plates, but only if healthcare organizations can find a secure enough solution. Confidential Computing can put that solution within reach.
  

• Automated SOAP notes: Mental health professionals face a similar challenge when it comes to the SOAP (Subjective, Objective, Assessment, Plan) notes they take to document sessions and recommendations for their patients. Confidential Computing can be a good option here too, providing a HIPAA-compliant option for collecting and using SOAP notes more effectively.
  

• Pre-LLM PHI redaction: One of the most important steps in achieving HIPAA-compliant cloud AI is removing all protected health information (PHI) before you provide it to an LLM (large language model). This is something AI can potentially help with, as long as you’re able to do it in a secure enough environment. A Google Cloud TEE can be the answer.
  
• Multi-hospital research: Hospital research can potentially go much further if different hospitals and researchers can find a secure way to share their findings. You can’t share raw patient records without risking compliance issues, but you can use a solution like Google Cloud’s Confidential Space to create a shared “clean room” to load all the sensitive data from multiple hospital sources in one place for analysis, without revealing anything private to the wrong parties.

AI has a lot of potential to help solve common problems in the healthcare industry, but only if you use a solution like Confidential Computing to tackle the confidentiality and compliance issues.

Promevo x IntelliTek Health Partnership

For any healthcare organization that’s interested in embracing AI, but worried about getting the security side of things right, working with a skilled partner can be just what you need to get started. Promevo, a Google Cloud partner with extensive experience in helping organizations launch AI projects, just teamed up with IntelliTek Health, a leader in AI-powered healthcare solutions.

Combine IntelliTek Health’s experience in the healthcare space with Promevo’s experience with Google Cloud, AI app development, and cloud migration and you get the best of all worlds. We can help you establish HIPAA-compliant cloud AI projects that solve your business problems, while maintaining the level of security your patients deserve (and your industry requires).

For a more detailed look at how we can help solve your particular problems, set up a call.