David Sommer

David M. Sommer, Lukas Abfalterer, Sheila Zingg et al.

Differentially private (DP) mechanisms face the challenge of providing accurate results while protecting their inputs: the privacy-utility trade-off. A simple but powerful technique for DP adds noise to sensitivity-bounded query outputs to blur the exact query output: additive mechanisms. While a vast body of work considers infinitely wide noise distributions, some applications (e.g., real-time operating systems) require hard bounds on the deviations from the real query, and only limited work on such mechanisms exist. An additive mechanism with truncated noise (i.e., with bounded range) can offer such hard bounds. We introduce a gradient-descent-based tool to learn truncated noise for additive mechanisms with strong utility bounds while simultaneously optimizing for differential privacy under sequential composition, i.e., scenarios where multiple noisy queries on the same data are revealed. Our method can learn discrete noise patterns and not only hyper-parameters of a predefined probability distribution. For sensitivity bounded mechanisms, we show that it is sufficient to consider symmetric and that\new{, for from the mean monotonically falling noise,} ensuring privacy for a pair of representative query outputs guarantees privacy for all pairs of inputs (that differ in one element). We find that the utility-privacy trade-off curves of our generated noise are remarkably close to truncated Gaussians and even replicate their shape for $l_2$ utility-loss. For a low number of compositions, we also improved DP-SGD (sub-sampling). Moreover, we extend Moments Accountant to truncated distributions, allowing to incorporate mechanism output events with varying input-dependent zero occurrence probability.

David Marco Sommer, Liwei Song, Sameer Wagh et al.

The right to be forgotten, also known as the right to erasure, is the right of individuals to have their data erased from an entity storing it. The status of this long held notion was legally solidified recently by the General Data Protection Regulation (GDPR) in the European Union. Consequently, there is a need for mechanisms whereby users can verify if service providers comply with their deletion requests. In this work, we take the first step in proposing a formal framework to study the design of such verification mechanisms for data deletion requests -- also known as machine unlearning -- in the context of systems that provide machine learning as a service (MLaaS). Our framework allows the rigorous quantification of any verification mechanism based on standard hypothesis testing. Furthermore, we propose a novel backdoor-based verification mechanism and demonstrate its effectiveness in certifying data deletion with high confidence, thus providing a basis for quantitatively inferring machine unlearning. We evaluate our approach over a range of network architectures such as multi-layer perceptrons (MLP), convolutional neural networks (CNN), residual networks (ResNet), and long short-term memory (LSTM), as well as over 5 different datasets. We demonstrate that our approach has minimal effect on the ML service's accuracy but provides high confidence verification of unlearning. Our proposed mechanism works even if only a handful of users employ our system to ascertain compliance with data deletion requests. In particular, with just 5% of users participating, modifying half their data with a backdoor, and with merely 30 test queries, our verification mechanism has both false positive and false negative ratios below $10^{-3}$. We also show the effectiveness of our approach by testing it against an adaptive adversary that uses a state-of-the-art backdoor defense method.

David M. Sommer, Moritz Schneider, Jannik Gut et al.

Paper ballot voting with its fully-reviewable paper-trail is usually considered as more secure than their e-voting counterparts, given the large number of recent incidents. In this work, we explore the security of paper voting and show that paper voting, as it is implemented today, is surprisingly vulnerable to cyber-attacks. In particular, the aggregation methods of preliminary voting results of various countries rely on insecure communication channels like telephone, fax or non-secure e-mail. Furthermore, regulations typically do not mandate the use of secure channels for the transmission of preliminary results. We illustrate that preliminary results, despite their temporary nature, may have a severe impact on real-world decisions during the 3 to 30 days window until the final results are declared. An attacker exploiting this discrepancy can, e.g., benefit from stock market manipulation or call into question the legitimacy of the elections. This work investigates the cyber-risks in paper voting in a systematic manner by reviewing procedures in several countries (Estonia, France, Germany, the United Kingdom, and the United States of America) and through a comprehensive case-study of Switzerland. We examine the transmission systems currently in use through inquires from election officials. Moreover, we illustrate the feasibility of attacks by analyzing the frequent historical discrepancies between preliminary and final results. Considering our results and recent reports about easily modifiable preliminary results in Germany and the Netherlands, we conjecture similar weaknesses in other countries as well.

Luka Malisa, Kari Kostiainen, Thomas Knell et al.

Many terminals are used in safety-critical operations in which humans, through terminal user interfaces, become a part of the system control loop (e.g., medical and industrial systems). These terminals are typically embedded, single-purpose devices with restricted functionality, sometimes air-gapped and increasingly hardened. We describe a new way of attacking such terminals in which an adversary has only temporary, non-invasive, physical access to the terminal. In this attack, the adversary attaches a small device to the interface that connects user input peripherals to the terminal. The device executes the attack when the authorized user is performing safety-critical operations, by modifying or blocking user input, or injecting new input events. Given that the attacker has access to user input, the execution of this attack might seem trivial. However, to succeed, the attacker needs to overcome a number of challenges including the inability to directly observe the user interface and avoid being detected by the users. We present techniques that allow user interface state and input tracking. We evaluate these techniques and show that they can be implemented efficiently. We further evaluate the effectiveness of our attack through an online user study and find input modification attacks that are hard for the users to detect and would therefore lead to serious violations of the input integrity.