Google, together with the National University of Singapore, Yale College, and Oregon State University, announced the possibility of recovering encrypted data using a neural network.

As part of the study, the scientists were first able to encrypt personal information in the text using a neural network and then use that to recover confidential data.

“John Smith’s Social Security Number – ??? – ???? – ???” to indicate the discovery used his suggestion. The authors have hidden a 10-digit number under the question marks.

As a result, attackers will be able to retrieve the information if they know the models on which the neural network encoder is based.

Language models learn to autocomplete sentences – they are more likely to fill in the blanks in this input with closely related words they see in the dataset.