Research at Sherpa.ai
Our research teams investigate how privacy and security shape the development of Artificial Intelligence, ensuring that its growing capabilities create a positive impact for companies, organizations, and society as a whole.
May 18th 2026
RESEARCH PAPER
PUBLISHED
Towards the Next Frontier of LLMs: Training on Private Data
The recent rise of Large Language Models (LLMs) has been driven by one foundational resource: public data. Books, websites, academic papers, code repositories, and open internet content have enabled AI systems to acquire broad linguistic and reasoning capabilities at unprecedented scale.
The world’s most valuable knowledge is not publicly accessible. It exists inside hospitals, banks, governments, industrial systems, telecommunications operators, and enterprise infrastructures. These environments contain patient histories, operational workflows, financial decisions, internal communications, cybersecurity events, and institutional expertise accumulated over decades.
April 19th 2026
RESEARCH PAPER
PUBLISHED
Sherpa.ai Privacy-Preserving Multi-Party Entity Alignment without Intersection Disclosure for Noisy Identifiers
Artificial intelligence is entering a phase where collaboration is no longer optional, but essential. The most valuable insights today emerge not from isolated datasets, but from the combination of multiple, distributed sources across organizations, industries, and borders. Yet this opportunity comes with a fundamental constraint: data cannot move.
Privacy regulations, competitive boundaries, and security risks make traditional data centralization not only impractical, but often impossible. In this context, the paper “Sherpa.ai Privacy-Preserving Multi-Party Entity Alignment without Intersection Disclosure for Noisy Identifiers” addresses one of the most critical and least solved challenges in modern AI: how to align and combine data across organizations without ever exposing the data itself.
December 19th 2025
RESEARCH PAPER
PUBLISHED
Training Together, Diagnosing Better: Federated Learning as a New Paradigm for Medical AI
National Institute of Health (NIH) | University College London (UCL) | Sherpa.ai
Rare disease diagnosis is fundamentally constrained by fragmented data, strict privacy regulations, and the scarcity of representative patient cohorts. This paper demonstrates how Federated Learning (FL) overcomes these barriers by enabling multiple clinical institutions to collaboratively train machine-learning models without sharing sensitive patient data. Applied to collagen VI–related dystrophies, the approach delivers significantly improved diagnostic accuracy and generalization across sites, achieving an F1-score of 0.82 compared to isolated institutional models. Beyond performance gains, this work establishes a scalable, privacy-preserving paradigm for medical AI, one that transforms collaboration itself into a diagnostic advantage.
November 14th 2025
RESEARCH PAPER
PUBLISHED
A global challenge: diagnosing pediatric pneumonia without sharing data
This paper presents a groundbreaking Federated Learning approach that transforms how pediatric pneumonia is detected across globally distributed healthcare systems. Leveraging the Sherpa.ai FL platform, it enables hospitals, pediatric clinics, and research centers to jointly train advanced AI models on chest imaging and clinical data without ever transferring or exposing sensitive patient information. The framework achieves a significan improvement in diagnostic accuracy while maintaining strict compliance with healthcare privacy regulations. This innovation marks a major step toward more accurate, equitable, and scalable pediatric care—making it possible to deliver earlier, data-driven pneumonia diagnosis for children worldwide, without compromising their privacy.
November 3rd 2025
RESEARCH PAPER
PUBLISHED
Federated Cyber Defense: Privacy-Preserving Ransomware Detection Across Distributed Systems
This paper presents a groundbreaking Federated Learning approach that revolutionizes how ransomware is detected across globally distributed systems. Leveraging the Sherpa.ai FL platform, it empowers cybersecurity companies, hardware manufacturers, and software providers to jointly train advanced AI models without ever moving or exposing private data. The framework achieves a 9% improvement in detection accuracy while maintaining full compliance with privacy regulations. This innovation marks a major leap toward secure, sustainable, and large-scale cyber defense, enabling proactive threat detection across millions of protected devices worldwide.
October 19th 2025
RESEARCH PAPER
PUBLISHED
Reducing by 99% the Communications of a Single Training
This paper presents a novel Federated Learning framework that enables secure, collaborative model training across multiple organizations without sharing raw data. By optimizing communication efficiency, it reduces network exchanges by up to 99%, dramatically enhancing data privacy and security. The proposed architecture improves scalability, lowers energy consumption, and achieves state-of-the-art performance across real-world applications in healthcare, finance, manufacturing, cybersecurity, aerospace and defense.
October 8th 2025
PATENT
GRANTED
From Voice to Presence: The Patent That Redefines the Future of Digital Assistants: U.S. Patent US 11,405,742 B2
In an era where large language models (LLMs) are redefining human–machine interaction, natural language has become our most intuitive interface — yet it’s only the beginning. Voice and text assistants still operate in flat, two-dimensional confines, missing the non-verbal depth that shapes over 70% of human communication. Sherpa.ai’s U.S. Patent 11,405,742 B2 introduces a 3D-audio assistant that projects its voice in space — whispering, alerting, or guiding naturally — transforming AI from a voice in a box into a real, spatial presence.
September 20th 2025
PATENT
GRANTED
System and method of integrating remote services: US Patent US8745194B2
This invention defines a universal AI platform that connects and automates every digital service through natural language. By combining a management core, dialogue engine, and multimodal interfaces with API-based connectors, it integrates banking, commerce, healthcare, education, mobility, entertainment, and smart environments. The result is a single intelligent system that enables fluid, secure, and human-centric interaction across all digital services in real time.