IBM says it has successfully carried out an experiment with a large bank to run machine-learning operations using homomorphic encryption, a technique that allows users to perform computations on data without having to decrypt it first.

“Until we worked with a bank, most of the tests and everything that we did was with synthetic data,” says Flavio Bergamaschi, the senior research scientist at IBM who lead the technical part of the project. “This time we worked with real financial data [and] we