Subverting Privacy-Preserving GANs: Hiding Secrets in Sanitized Images

Kang Liu, Benjamin Tan, Siddharth Garg

Introduction

Privacy-Preserving Generative Adversarial Networks (PP-GANs) have emerged as a promising solution for data sanitization. These tools aim to transform sensitive datasets—such as facial images—into versions that retain utility (e.g., facial expressions) while removing private attributes (e.g., identity). However, the security of these systems often rests on empirical privacy audits: if a secondary "discriminator" network cannot identify the private attribute in the sanitized output, the tool is deemed secure.

The threat model of a subverted PP-GAN. While the end user's privacy check passes, the malicious provider can still recover the original identity from the sanitized output.

This paper demonstrates that such empirical checks are fundamentally flawed. We show that a malicious model provider can design a "subverted" PP-GAN that acts as a Trojan horse. This adversarial model produces outputs that appear perfectly sanitized to both human eyes and automated discriminators, yet contain hidden, recoverable secrets using steganographic techniques.

The Adversarial PP-GAN Framework

The proposed attack, built upon the PPRL-VGAN architecture, integrates secret extraction and embedding directly into the neural network's weights. The process involves three distinct stages:

  1. Secret Extraction: The GAN's encoder is modified with an additional fully connected layer ($FC\_0$). This layer is trained to capture the "secret"—either a discrete user ID (Scheme 1) or a compressed latent representation of the entire input image (Scheme 2).
  2. Secret Embedding: The extracted secret is hidden within the frequency domain of the sanitized image. By implementing a Discrete Cosine Transform (DCT) as a differentiable neural layer, the adversary can manipulate the Least Significant Bits (LSBs) of middle-frequency coefficients. This ensures the hidden data is robust and invisible to the human eye.
  3. Innocuous Architecture: To evade detection by users inspecting the model's structure, the adversarial paths are merged into a standard convolutional architecture, making the subverted model look like a standard multi-layer network.
Detailed architecture of the subverted PP-GAN. The model extracts a secret via FC_0 and embeds it into the sanitized image using a Secret Generation Network and DCT/Inverse-DCT layers.

Experimental Evaluation

The attack was evaluated on two facial expression datasets: FERG (stylized characters) and MUG (real human faces). The results confirm that the subverted models successfully balance three competing goals:

  • Utility Preservation: The models maintained 100% accuracy in facial expression recognition, matching benign performance.
  • Privacy Check Evasion: The subverted images passed both "weak" and "strong" privacy audits. Discriminators trained to identify users from the sanitized images performed no better than they did on honestly sanitized data (e.g., ~30% accuracy on FERG).
  • Secret Recovery: The adversary achieved near-perfect recovery. In Scheme 1, user IDs were recovered with up to 100% accuracy. In Scheme 2, the adversary reconstructed the original private images with a Mean Squared Error (MSE) of $3.36 \times 10^{-4}$, producing visually near-identical replicas.

Conclusion

The success of these subversion attacks highlights a critical weakness in current privacy-preserving machine learning: the reliance on empirical "cat-and-mouse" games. Passing a privacy check does not guarantee the absence of sensitive information; it only indicates the failure of a specific detection method. To build truly secure systems, the community must move toward formal privacy frameworks, such as Differential Privacy, and exercise extreme caution when deploying third-party "black-box" sanitization tools.

Summary updated

Abstract

Unprecedented data collection and sharing have exacerbated privacy concerns and led to increasing interest in privacy-preserving tools that remove sensitive attributes from images while maintaining useful information for other tasks. Currently, state-of-the-art approaches use privacy-preserving generative adversarial networks (PP-GANs) for this purpose, for instance, to enable reliable facial expression recognition without leaking users' identity. However, PP-GANs do not offer formal proofs of privacy and instead rely on experimentally measuring information leakage using classification accuracy on the sensitive attributes of deep learning (DL)-based discriminators. In this work, we question the rigor of such checks by subverting existing privacy-preserving GANs for facial expression recognition. We show that it is possible to hide the sensitive identification data in the sanitized output images of such PP-GANs for later extraction, which can even allow for reconstruction of the entire input images, while satisfying privacy checks. We demonstrate our approach via a PP-GAN-based architecture and provide qualitative and quantitative evaluations using two public datasets. Our experimental results raise fundamental questions about the need for more rigorous privacy checks of PP-GANs, and we provide insights into the social impact of these.
BibTeX
@inproceedings{liu2021subverting,
  author = {Kang Liu and Benjamin Tan and Siddharth Garg},
  title = {{Subverting Privacy-Preserving GANs: Hiding Secrets in Sanitized Images}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2021},
  doi = {10.1609/aaai.v35i17.17743},
}
Figure