The field of secure data analysis and private inference is rapidly advancing, with a focus on developing innovative methods for analyzing encrypted data and generating private insights. Recent developments have led to the creation of secure architectures for analyzing messaging trends in end-to-end encrypted communications, enabling investigative journalism workflows while maintaining user privacy. Additionally, there have been significant improvements in fully homomorphic encryption, allowing for efficient and accurate computation over encrypted data. Noteworthy papers in this area include:

• Synopsis, which introduces a secure architecture for analyzing messaging trends in consensually-donated end-to-end encrypted messages.
• Hermes, which enables high-performance homomorphically encrypted vector databases, bringing fully homomorphic encryption from cryptographic promise to practical reality.
• Clustering and Median Aggregation Improve Differentially Private Inference, which presents a new algorithm for aggregating next token statistics by privately computing medians, allowing for high-quality synthetic data generation at lower privacy cost.
• Urania, which introduces a novel framework for generating insights about LLM chatbot interactions with rigorous differential privacy guarantees, providing end-to-end privacy protection.