An algorithm that hides sensitive information so well that it is impossible to tell that anything has been hidden.
‘A new family of steganography algorithms that have perfect security guarantees.’
Researchers have made a big step forward in secure communications by making an algorithm that hides sensitive information so well that it is impossible to tell that anything has been hidden.
The team, which was led by the University of Oxford and worked closely with Carnegie Mellon University, thinks that this method may soon be used widely in digital human communications, such as social media and private messaging.
In particular, being able to send information that is completely safe could help vulnerable groups like dissidents, investigative journalists, and people who work in humanitarian aid.
The method may be used in the context of steganography, the art of concealing important information inside seemingly unrelated material. Steganography is different from cryptography because sensitive information is hidden in a way that makes it hard to tell that it has been hidden. A piece of Shakespeare may be buried in an AI-created picture of a cat.
Even though steganography has been studied for more than 25 years, most of the current methods are not very secure. This means that people who use these methods risk being caught. This is due to the fact that traditional steganography algorithms used a subtle shift in the normal content’s distribution.
To solve this problem, the research team used recent advances in information theory. Specifically, they used minimum entropy coupling, which lets you connect two data distributions so that their mutual information is maximized but their individual distributions are kept.
Hence, the new approach does not detect any statistically significant disparity between the distributions of benign and secretly encoded items.
Many models that create AI-generated material were used to evaluate the method, including the open-source language model GPT-2 and the text-to-speech converter WAVE-RNN.
The new algorithm demonstrated up to 40% greater encoding efficiency than earlier steganography techniques across a range of applications in addition to being completely secure, allowing for the hiding of more information within a given quantity of data.
Steganography’s advantages for data reduction and storage may make it enticing even if full security is not needed.
The study team has applied for a patent on the method, but they want to make it available to the public under a free license for ethical non-commercial usage.
This covers usage for scholarly and humanitarian purposes, as well as security assessments conducted by reputable third parties. The researchers have shared an ineffective implementation of their methodology on Github and submitted their work as a preprint paper on arXiv.
And in May, 2023, at the world’s most prestigious artificial intelligence conference, the International Conference on Learning Representations, they will reveal the new algorithm they developed.
Products like ChatGPT, Snapchat AI-stickers, and TikTok video filters have contributed to the widespread use of AI-generated material in everyday human communication.
Hence, steganography may become more prevalent, since the sheer appearance of AI-generated material will no longer raise suspicion.
Dr. Christian Schroeder de Witt, co-lead author from the Department of Engineering Science at the University of Oxford, stated that their method has the potential to be utilized with any software that generates content automatically, such as probabilistic video filters or meme generators. This could be particularly useful for journalists and aid workers in countries where encryption is prohibited. However, caution should still be exercised by users since any encryption method could be vulnerable to side-channel attacks, such as the detection of a steganography app on a user’s phone.
Meanwhile, co-lead author Samuel Sokota from the Machine Learning Department at Carnegie Mellon University explained that their work’s primary contribution is demonstrating a profound connection between minimum entropy coupling and perfectly secure steganography. By leveraging this connection, “we introduce a new family of steganography algorithms that have perfect security guarantees.”
Source: 10.48550/arXiv.2210.14889
Image Credit: Getty
