Confidential AI and the Rise of Privacy Enhancing Technology

What Are PETs?
PETs make it possible to analyse and learn from data while reducing the risk of privacy breaches. They are used in everything from training AI models to joint research between companies.

Some common types include:
– Differential Privacy: Adds random noise to protect individuals in large datasets.
– Federated Learning: Trains models across multiple locations or devices without needing to centralise the data.
– Homomorphic Encryption: Allows calculations on encrypted data, without first decrypting it.
– Secure Multi-Party Computation: Lets different parties compute a result together without revealing their individual inputs.
– Trusted Execution Environments (TEEs): Hardware-based secure areas that keep data safe during processing.

Each method helps balance the need for insights with the obligation to protect privacy.

PETS Innovation from AWS
Amazon Web Services is actively building PETs into its platform.

Three of their solutions stand out:
– AWS Nitro Enclaves create isolated computing zones that protect sensitive workloads, like handling encryption keys or personal data.
– Amazon SageMaker with Federated Learning supports training AI models using data from multiple locations, which is especially useful in healthcare and industrial environments.
– AWS Clean Rooms let multiple organisations analyse shared data without seeing the raw inputs from each other.

For example, a pharmaceutical company could work with hospitals in different countries to train drug response models. Using PETs like SageMaker and Clean Rooms, they can do this without moving or exposing any patient data.

Privacy and the Law
PETs are also helping companies meet their obligations under global privacy regulations such as:

– The General Data Protection Regulation (GDPR) in the European Union
– CCPA/CPRA in California
– The Australian Privacy Act
– The EU AI Act, which sets out rules for fairness, transparency and responsible data use

By using PETs from the start, organisations align with privacy-by-design principles and reduce their legal and reputational risks.

Challenges to PET Adoption
While promising, PETs are not without challenges.

– Performance: Some approaches, like homomorphic encryption, are still slower than conventional processing.
– Complexity: Integrating multiple PETs into a single workflow can be technically difficult.
– Accuracy: Some methods, such as differential privacy, can affect the quality of results.

Because of these issues, many organisations use PETs selectively, based on the sensitivity of the data and the regulatory requirements they face.

What’s Ahead for Confidential AI
The next wave of PET adoption will likely be driven by:

– Easy to use tools for developers
– Cloud based services offering PETs as part of their standard products
– International standards and frameworks to guide implementation
– Prebuilt PET modules that can be added to existing data pipelines

Confidential AI is shifting from concept to common practice, particularly in sectors that handle sensitive information.

In Closing
Confidential AI is not just about ticking compliance boxes. It is about creating space for innovation while building trust.

Privacy Enhancing Technologies are giving organisations the ability to collaborate securely, train AI responsibly, and unlock the value of data without compromising what matters most, people’s privacy.

Those who adopt Confidential AI early are likely to differentiate themselves and lead in a world that is increasingly driven by data.

References
– Amazon Web Services (2024) AWS Clean Rooms. Available at: https://aws.amazon.com/clean-rooms/ (Accessed: 2 July 2025)
– OECD (2023) Sharing Trustworthy AI Models with Privacy-Enhancing Technologies. Available at: https://www.oecd.org/en/publications/sharing-trustworthy-ai-models-with-privacy-enhancing-technologies_a266160b-en.html (Accessed: 2 July 2025)
– World Economic Forum (2023) The Impact of PETs on Business and Society. Available at: https://www.weforum.org/stories/2023/10/the-impact-of-privacy-enhancing-technologies-pet-on-business-individuals-and-society/ (Accessed: 2 July 2025)
– Marketingscoop (2024) Privacy Enhancing Technologies: The Future of Secure AI. Available at: https://www.marketingscoop.com/ai/privacy-enhancing-technologies/ (Accessed: 2 July 2025)