New encryption method uses AI-generated content to hide info in plain sight

With the growth of ChatGPT and other generative AI content online, an unintended benefit just might be more secure digital communications.

A new encryption algorithm has just been published on the preprint server arXive that describes a new steganographic technique that relies on AI-generated images to with a hidden image encoded within it in a way that hides the fact that something is even being hidden.

Steganography isn't a new idea, but existing steganographic techniques leave telltale artifacts in their information distribution that can be detected, cluing the observer into the fact that something is being hidden within the content.

With AI-generated content like images, though, the new algorithm uses a technique called minimum entropy coupling to maximize the overlap between the information distribution of the innocuous content and the content one wants to encode to make it so that there is no statistical difference between the two distributions.

Essentially, you can use the algorithm to AI-generate content that perfectly masks the information you are trying to hide, and given the rise of AI content online, such content — like DALL-E 2 imagery — wouldn't in itself be a cause for suspicion. The encoded information is easily accessible if you have the key or cipher for the information you are looking to extract from the image, however.

"The main contribution of the work is showing a deep connection between a problem called minimum entropy coupling and perfectly secure steganography," co-lead author Samuel Sokota, from Carnegie Mellon University's Machine Learning Department, said in a statement. "By leveraging this connection, we introduce a new family of steganography algorithms that have perfect security guarantees."

What's more, the new algorithm also shows up to 40% better efficiency when encoding information, so even if the heightened level of security isn't warranted, the algorithm might still prove useful for its data compression and storage benefits.

"Our method can be applied to any software that automatically generates content, for instance, probabilistic video filters, or meme generators," said co-lead author Dr. Christian Schroeder de Witt, of Oxford's Department of Engineering Science. "This could be very valuable, for instance, for journalists and aid workers in countries where the act of encryption is illegal. 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."

Rise of AI-generated content makes a perfect smokescreen to hide sensitive info

The researchers have applied for a patent for the algorithm and have already published an inefficient implementation of the algorithm on the popular code repository Github as open source code.

Even though the algorithm will likely receive patent protection, the researchers said they intend to make the algorithm freely available through a third-party license for non-commercial and responsible use, such as academic and humanitarian purposes.

Such an encryption scheme is important, especially in places where encryption might be illegal or prohibited to limit free expression or stifle dissent, and as more and more AI-generated content is produced and disseminated, it will become increasingly impossible to even know where to look for encoded information, making the security of the new method that much stronger.