Provably Secure Generative Linguistic Steganography

Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 3046–3055 August 1–6, 2021. ©2021 Association for Computational Linguistics 3046 Provably Secure Generative Linguistic Steganography Siyu Zhang, Zhongliang Yang, Jinshuai Yang, Yongfeng Huang Department of Electronic Engineering, Tsinghua University, Beijing 100084, China {zhangsiy19, yjs20}@mails.tsinghua.edu.cn yangzl15@tsinghua.org.cn yfhuang@mail.tsinghua.edu.cn Abstract Generative linguistic steganography mainly utilized language models and applied stegano- graphic sampling (stegosampling) to gener- ate high-security steganographic text (stego- text). However, previous methods generally lead to statistical differences between the con- ditional probability distributions of stegotext and natural text, which brings about security risks. In this paper, to further ensure secu- rity, we present a novel provably secure gener- ative linguistic steganographic method ADG, which recursively embeds secret information by Adaptive Dynamic Grouping of tokens ac- cording to their probability given by an off- the-shelf language model. We not only prove the security of ADG mathematically, but also conduct extensive experiments on three pub- lic corpora to further verify its imperceptibil- ity. The experimental results reveal that the proposed method is able to generate stegotext with nearly perfect security. 1 Introduction Steganography is the technology of hiding secret in- formation within an innocent natural carrier (such as image (Hussain et al., 2018), audio (Mishra et al., 2018), video (Liu et al., 2019), text (Krishnan et al., 2017), etc) in order to avoid eavesdropping. Steganography differs from cryptography in that cryptography only conceals the content of secret information, whereas steganography even conceals its very existence, which makes it more secure and reliable in some scenarios (Anderson and Petitco- las, 1998). Natural language is suitable as a carrier of steganography by virtue of its high robustness in transmission (Ziegler et al., 2019). Unlike digi- tal images or digital audio which is sensitive to distortions like compression, cropping, blurring or pixel-wise dropout, text can usually be transmitted losslessly through different kinds of public chan- nels. Nevertheless, text generally has low entropy and lacks sufficient redundancy for information hiding (Sharma et al., 2016), which often results in low embedding capacity of linguistic steganog- raphy. For example, in traditional modification- based methods (such as synonym substitution (Xi- ang et al., 2014, 2018) and spelling transformation (Shirali-Shahreza, 2008)), where secret informa- tion is encoded by slightly modifying an existing covertext, the options for modification can be very limited to keep the text fluent enough so as not to arouse suspicions. In recent years, powered by the advanced tech- nology of deep learning and natural language pro- cessing, language models based on neural networks have made significant progress in generating flu- ent text (Radford et al., 2019; Brown et al., 2020), which bring new vitality to linguistic steganogra- phy and facilitate the investigation of generation- based methods (Fang et al., 2017; Yang et al., 2018a; Dai and Cai, 2019; Ziegler et al., 2019; Yang et al., 2020a; Zhou et al., 2021). The gener- ative linguistic steganography directly transform secret information into innocuous-looking stegano- graphic text (stegotext) without any covertext. Us- ing an off-the-shelf language model, secret infor- mation can …

Content truncated to 500 words