November 2025 Summaries

Quantum computing presents significant opportunities but also poses a substantial threat to current cryptographic methods, as future quantum machines could potentially decrypt data secured by traditional encryption algorithms like RSA and ECC. This risk is not just a future concern; adversaries are already collecting encrypted data with the intention of decrypting it once quantum technology becomes viable. Duality Technologies addresses this issue by integrating post-quantum cryptography (PQC) into Fully Homomorphic Encryption (FHE), creating a quantum-resilient system that secures data both in transit and in use. This approach, termed "Zero-Footprint Intelligence," allows for encrypted queries on external datasets while keeping investigative intent and sensitive metadata private. Duality emphasizes the urgency of adopting post-quantum FHE to secure data now, ensuring its confidentiality against future quantum decryption capabilities. This technology not only safeguards government analytics but also establishes a foundation for Confidential AI, capable of withstanding emerging cryptographic threats, thereby ensuring long-term data protection.

As global data privacy regulations such as the EU's GDPR, California's CPRA, India's DPDP Act, and the impending EU AI Act become increasingly stringent, AI systems face new challenges and opportunities in handling sensitive data. These regulations emphasize the necessity for robust privacy measures during data processing, not just storage, compelling enterprises to adopt Privacy-Enhancing Technologies (PETs) like federated learning, confidential computing, and homomorphic encryption. Such technologies enable secure computation on sensitive data without exposure, ensuring compliance and fostering trust while allowing cross-border collaboration. As fines for non-compliance grow, the regulatory landscape signals a shift towards architectures that inherently protect data in use, urging AI systems to demonstrate transparency, auditability, and privacy by design in their operations. This evolving framework demands that AI strategies adapt to ensure that models can function without direct access to raw data, thereby aligning with the emerging global data governance standards.

Pharmaceutical companies are increasingly turning to AI-driven prescreening models, powered by privacy-enhancing technologies (PETs), to address delays in clinical trial enrollment due to the challenges of patient prescreening. These models can predict a patient's likelihood of meeting trial eligibility criteria without moving or exposing sensitive data, thanks to technologies like Trusted Execution Environments (TEEs) and Federated Inference. By keeping patient data secure and running models locally, PETs enable faster and more efficient prescreening processes, reducing screen-failure rates and shortening the time between referral and enrollment. This approach is transforming trial operations from a reactive to a proactive model, allowing for quicker trial startups and enhanced patient diversity while ensuring compliance with data privacy regulations. A global pharmaceutical company has already demonstrated the effectiveness of this approach by implementing PETs to enhance biomarker prediction across diverse data sources without compromising patient privacy or model integrity, marking a significant shift towards more efficient and trusted clinical trial processes.

The UK government is presented with a unique opportunity to lead in the development of responsible, high-value AI by establishing a cross-economy AI Growth Lab that addresses the challenges of data access, regulatory fragmentation, and organisational hesitation. The proposal emphasizes the need for centralized coordination across regulators to provide clear guidance and reduce legal risks, as well as the importance of focusing on commercially relevant and scalable AI applications that rely on cross-sector datasets. By embedding privacy-enhancing technologies and secure infrastructure within the Lab, it aims to facilitate real-world testing without compromising data security. The success of this initiative depends on transparent governance, real-time monitoring, and central government ownership to ensure consistent regulatory guidance across various sectors, ultimately setting a global standard for deploying AI responsibly on sensitive data.

As federated learning evolves from research to real-world applications, the need for interoperability has become vital, prompting organizations to use open-source frameworks like Flower, known for its simplicity and flexibility. However, challenges arise when operationalizing these workloads securely at scale. Duality addresses this issue by integrating native support for Flower within its platform, allowing organizations to transition their Flower-based federated learning code into a secure, production-ready environment without rewriting it. This integration combines Duality's enterprise-grade features such as governance, policy control, privacy-enhancing technologies, scalable orchestration, and compliance with Flower's developer-friendly framework, ensuring that sensitive data remains protected while facilitating a seamless path from research to deployment. Duality serves as a bridge between research-focused frameworks like Flower and enterprise-ready systems like NVIDIA FLARE, unifying governance and policy enforcement across federated learning ecosystems. This development enables organizations to achieve both open-source agility and enterprise compliance, thereby accelerating the journey from conceptual models to scalable, secure deployment.