San Francisco / Leuven — October 20, 2026 Belfort today released EML, a fast and practical implementation of encrypted computing that runs your own PyTorch model. EML combines fast GPU acceleration with tooling to translate AI models to the encrypted domain. You provide it with your neural network model and it provides the framework to allow encrypted inference using this model. Sign up for the beta release at eml.belfortlabs.com. 

EML: fast & practical encrypted PyTorch

At Belfort, we believe that the next era of computing will be encrypted. As AI matures, data security has become the number one concern for enterprises. Making this new encrypted computing paradigm work takes two things: speed, and low complexity. Our ResNet demo in July was a first step towards making computations fast. Today, we deliver on the complexity front with EML, a package that is simple to use. It combines a compiler toolchain, built around Google HEIR, with the fastest GPU library available, running on Nvidia hardware. 

Belfort’s CTO Michiel Van Beirendonck explains:

“We are pushing on two frontiers, speed and ease of use. Speed was the harder problem for a long time. Now the harder problem is that only a few dozen people in the world can write an encrypted model by hand. EML removes that constraint. ” 

Built for the real world with early partners

A common critique of encrypted computing is that it is still a research topic. Belfort defies this expectation and has been working with early partners such LG, NXP and Google to better understand the major roadblocks when  using encrypted computing.

‍Laurens De Poorter, CEO at Belfort, states:

“Our partners never asked us for a faster benchmark. They asked to help them with their encrypted workloads. We speak with them weekly, and EML is the answer to that request.”

Our valued partner Google has been a cornerstone of this effort with the HEIR compiler. Together, we have added necessary extensions to run machine learning workloads in the encrypted domain.

The fastest encrypted computing in the world

Under the hood, Belfort uses a hyper efficient GPU library. Focused on different machine learning, Belfort outperforms similar offerings of competitors by a factor 3 or more. We are still improving our implementation to achieve speedups across different model architectures.

What’s next and availability

EML is currently available in beta through our waiting list. Individuals and organisations interested in running it, can sign up to eml.belfortlabs.com. 

‍

About Belfort

Belfort is building the foundation for trustless intelligence. Belfort builds encrypted computing (FHE) infrastructure so sensitive data can be processed without ever being decrypted. Spun out of KU Leuven’s COSIC lab, its hardware-software co-design makes this practical at scale. The company has offices in San Francisco, USA, and Leuven, Belgium. https://belfortlabs.com/

‍