AI's Can Produce Secure Steganographic Images

Via https://www.schneier.com/

https://www.quantamagazine.org/secret-messages-can-hide-in-ai-generated-media-20230518/

https://arxiv.org/abs/2210.14889    Tech Paper

New research suggests that AIs can produce perfectly secure steganographic images:

Abstract: Steganography is the practice of encoding secret information into innocuous content in such a manner that an adversarial third party would not realize that there is hidden meaning. While this problem has classically been studied in security literature, recent advances in generative models have led to a shared interest among security and machine learning researchers in developing scalable steganography techniques. In this work, we show that a steganography procedure is perfectly secure under Cachin (1998)’s information theoretic-model of steganography if and only if it is induced by a coupling. Furthermore, we show that, among perfectly secure procedures, a procedure is maximally efficient if and only if it is induced by a minimum entropy coupling. These insights yield what are, to the best of our knowledge, the first steganography algorithms to achieve perfect security guarantees with non-trivial efficiency; additionally, these algorithms are highly scalable. To provide empirical validation, we compare a minimum entropy coupling-based approach to three modern baselines—arithmetic coding, Meteor, and adaptive dynamic grouping—using GPT-2, WaveRNN, and Image Transformer as communication channels. We find that the minimum entropy coupling-based approach achieves superior encoding efficiency, despite its stronger security constraints. In aggregate, these results suggest that it may be natural to view information-theoretic steganography through the lens of minimum entropy coupling. ... '

China Espionage via GE?

Note use of Steganography, hiding data.  an old but still used method for espionage.   I taught Chinese students at Columbia.

Industrial espionage: How China sneaks out America's technology secrets

By Nicholas Yon, in BBC News

It was an innocuous-looking photograph that turned out to be the downfall of Zheng Xiaoqing, a former employee with energy conglomerate General Electric Power.

According to a Department of Justice (DOJ) indictment, the US citizen hid confidential files stolen from his employers in the binary code of a digital photograph of a sunset, which Mr Zheng then mailed to himself.

It was a technique called steganography, a means of hiding a data file within the code of another data file. Mr Zheng utilised it on multiple occasions to take sensitive files from GE.

GE is a multinational conglomerate known for its work in the healthcare, energy and aerospace sectors, making everything from refrigerators to aircraft engines.

The information Zheng stole was related to the design and manufacture of gas and steam turbines, including turbine blades and turbine seals. Considered to be worth millions, it was sent to his accomplice in China. It would ultimately benefit the Chinese government, as well as China-based companies and universities.

Zheng was sentenced to two years in prison earlier this month. It is the latest in a series of similar cases prosecuted by US authorities. In November Chinese national Xu Yanjun, said to be a career spy, was sentenced to 20 years in prison for plotting to steal trade secrets from several US aviation and aerospace companies - including GE.  ... ' 

Hiding Malware in Artificial Neurons

 My areas of interest have always included machine learning, neural networks, steganography and security.    So I read this in interest. Not sure how it would work in practice.  Following up.

(When I say I am following up, I may or may not include findings in latter posts at my discretion.  I will if I think its particularly useful.   Do let me know if you have interest) 

ACM NEWS

Researchers Hid Malware Inside an AI's 'Neurons' And It Worked Scarily Well   July 23, 2021

The authors concluded that a 178MB AlexNet model can have up to 36.9MB of malware embedded into its structure without being detected using a technique called steganography. ... 

Neural networks could be the next frontier for malware campaigns as they become more widely used, according to a new study. 

According to the study, which was posted to the arXiv preprint server  on Monday, malware can be embedded directly into the artificial neurons that make up machine learning models in a way that keeps them from being detected. The neural network would even be able to continue performing its set tasks normally.

"As neural networks become more widely used, this method will be universal in delivering malware in the future," the authors, from the University of the Chinese Academy of Sciences, write.

View Full Article

Uncrackable Invisible Ink?

 Stuck me, since some of my earliest looks at code were in this realm.  Steganographic or hidden information.    Not combine it with AI based methods? 

An Uncrackable Combination of Invisible Ink, AI

American Chemical Society, May 5, 2021

Researchers have printed complexly encoded data using a carbon nanoparticle-based ink that can be read only by an artificial intelligence (AI) model when exposed to ultraviolet (UV) light. The researchers created the ‘invisible’ ink, which appears blue when exposed to UV light, using carbon nanoparticles from citric acid and cysteine. They then trained an AI model to identify symbols written in the ink and illuminated by UV light, and to use a special codebook to decode them. The model, which was tested using a combination of normal red ink and UV fluorescent ink, read the messages with 100% accuracy. The researchers said the algorithms potentially could be used for secure encryption with hundreds of unpredictable symbols because they can detect minute modifications in symbols.

Storing Data in Music

Intriguing idea, in theory not that hard.   Why might it be used?  A kind of Steganography?

Storing Data in Music 
ETH Zurich
By Fabio Bergamin

Researchers at ETH Zurich in Switzerland have developed a method for embedding data in music in a way that is imperceptible to the human ear, and transmitting it to a smartphone. The researchers found that under ideal conditions, the technique can transfer up to 400 bits per second without the average listener noticing. The researchers used the dominant notes in a piece of music, overlaying each of them with two marginally deeper and two marginally higher notes that are quieter than the dominant note. The team also used the harmonics of the strongest note, inserting slightly deeper and higher notes there as well. The data is stored in these additional notes. Said ETH’s Simon Tanner, “What we’re doing is embedding the data in the music itself; transmitting data from the loudspeaker to the mic.”  .... '