A team of researchers has developed a new method of hiding information inside innocuous-looking documents.

The technique itself – steganography – has been used in various forms for centuries; it differs from cryptography because the information has not been scrambled, just hidden. One early example would be the use of invisible ink to conceal a message in an ordinary-looking letter. In the computer age, so-called stego apps can be used to add secret data into image or video files by subtly altering individual pixels, or adding invisible text to the white spaces of text documents or to empty cells on spreadsheets.

Then there are hackers, who use steganography to hide malicious code – like scripts – in innocent-looking image files or macro-enabled text and Excel files, without triggering antivirus software. When the dodgy document is opened, the hidden script installs a small programme on the computer which can then download and install other malware. Steganography can also be used by attackers to exfiltrate stolen data without arousing suspicion because it is hidden in another document.

However, hiding lines of code in an image – like a JPEG – alters the underlying document, adding more data, and can result in a larger-than-usual file size. Therefore, if someone emails you a picture purporting to be of their cat, and the file size is multiple Megabytes even though the image is only 800×600 pixels, there might just be something untoward hiding in that file.

All that could be about to change though.

On Tuesday, researchers from the UK’s University of Oxford and Carnegie Mellon University in the US announced a new steganography algorithm that renders imperceptible any alterations to the underlying document in which data has been concealed.

The algorithm is a practical application of a theory called minimum entropy coupling, which as the name suggests, is a method for joining together two sets of information by combining their mutual data, but at the same time preserving them as individual data sets.

The team tested their algorithm by using AI-content generators – including the GPT-2 language model, and a text-to-speech programme called WAVE-RNN – as mules for hidden data. Not only did the algorithm prove perfectly secure, it also improved encoding efficiency by up to 40% compared to other steganography methods, meaning a larger volume of data could be concealed.

“Our method can be applied to any software that automatically generates content, for instance probabilistic video filters, or meme generators. This could be very valuable, for instance, for journalists and aid workers in countries where the act of encryption is illegal,” said a statement from co-lead author Dr Christian Schroeder de Witt, from Oxford University’s department of engineering science.

“However, users still need to exercise precaution as any encryption technique may be vulnerable to side-channel attacks such as detecting a steganography app on the user’s phone,” he added.

The researchers leaned heavily on the dissident/humanitarian angle, but as we all know, tools like this can be used for bad as well as good. Telecoms.com asked the researchers what would stop a hacker from exploiting this new algorithm to access corporate networks and steal data, and will update this story if and when they respond.

Meanwhile, the researchers have filed a patent for the algorithm, but intend to issue it under a free licence to third parties for non-commercial responsible use. This includes academic and humanitarian use, and trusted third-party security audits. The researchers have also published their work as a preprint paper on arXiv – a platform for distributing free scholarly articles – and released an open-source but inefficient implementation of their method on Github.

UPDATE – 16:30 7 March 23: We received the below response, from Oxford University’s Dr Christian Schroeder de Witt, to the following questions. “Is there a risk that minimum entropy coupling can be exploited by those who want to infect corporate computers with malware and steal sensitive data? How does IT security defend against these kinds of attacks when they have been rendered undetectable?”

“Any secure communications technology is dual-use (this has been publicly discussed in detail since the 90s Crypto Wars). While our method, iMEC, may help humanitarian aid workers and investigative journalists evade authoritarian surveillance, it may also be used in other contexts. By attaining perfect security properties, together with the increasing quality and proliferation of AI-generated content, iMEC may indicate that the race to detect steganography based on statistical analysis (steganalysis) of public communications alone may have been decided in favor of steganography. Law enforcement, however, has (and has in the past made use of) additional means of pursuing the illegal use of steganography, for example by detecting the presence of steganography software. While this may pose a challenge to lawmakers in the near term, this also means that authoritarian mass surveillance may become less feasible. Some actors such as the Signal Foundation believe that access to truly private communications is a precondition for a livable future. While this public discussion is ongoing, we, as responsible researchers, for now, have taken steps to control access to our production software.”

Get the latest news straight to your inbox. Register for the Telecoms.com newsletter here.