Helmholtz Center for Information Security
CISPA – Helmholtz-Zentrum für InformationssicherheitNot a member yet
3406 research outputs found
Sort by
Differential Analysis and Fingerprinting of ZombieLoads on Block Ciphers
Get PDFMicroarchitectural Data Sampling (MDS) enables to observe in-flight data that has recently been loaded or stored in shared short-time buffers on a physical CPU core. In-flight data sampled from line-fill buffers (LFBs) are also known as "ZombieLoads".
We present a new method that links the analysis of ZombieLoads to Differential Power Analysis (DPA) techniques and provides an alternative way to derive the secret key of block ciphers. This method compares observed ZombieLoads with predicted intermediate values that occur during cryptographic computations depending on a key hypothesis and known data. We validate this approach using an Advanced Encryption Standard (AES) software implementation.
Further, we provide a novel technique of cache line fingerprinting that reduces the superposition of ZombieLoads from different cache lines in the data sets resulting from an MDS attack. Thereby, this technique is helpful to reveal static secret data such as AES round keys which is shown in practice on an AES implementation
Towards Causal VQA: Revealing and Reducing Spurious Correlations by Invariant and Covariant Semantic Editing
No full textDespite significant success in Visual Question Answering (VQA), VQA models have been shown to be notoriously brittle to linguistic variations in the questions. Due to deficiencies in models and datasets, today's models often rely on correlations rather than predictions that are causal w.r.t. data.
In this paper, we propose a novel way to analyze and measure the robustness of the state of the art models w.r.t semantic visual variations as well as propose ways to make models more robust against spurious correlations.
Our method performs automated semantic image manipulations and tests for consistency in model predictions to quantify the model robustness as well as generate synthetic data to counter these problems.
We perform our analysis on three diverse, state of the art VQA models and diverse question types with a particular focus on challenging counting questions. In addition, we show that models can be made significantly more robust against inconsistent predictions using our edited data. Finally, we show that results also translate to real-world error cases of state of the art models, which results in improved overall performance
Mining Input Grammars from Dynamic Control Flow
Get PDFOne of the key properties of a program is its input specification. Having a formal input specification can be critical in fields such as vulnerability analysis, reverse engineering, software testing, clone detection, or refactoring. Unfortunately, accurate input specifications for typical programs are often unavailable or out of date.
In this paper, we present a general algorithm that takes a program and a small set of sample inputs and automatically infers a readable context-free grammar capturing the input language of the program. We infer the syntactic input structure only by observing access of input characters at different locations of the input parser. This works on all stack based recursive descent input parsers, including parser combinators, and works entirely without program specific heuristics. Our Mimid prototype produced accurate and readable grammars for a variety of evaluation subjects, including complex languages such as JSON, TinyC, and JavaScript
Artificial Fingerprinting for Generative Models: Rooting Deepfake Attribution in Training Data
No full textPhotorealistic image generation has reached a new level of quality due to the breakthroughs of generative adversarial networks (GANs). Yet, the dark side of such deepfakes, the malicious use of generated media, raises concerns about visual misinformation. While existing research work on deepfake detection demonstrates high accuracy, it is subject to advances in generation techniques and adversarial iterations on detection countermeasure techniques. Thus, we seek a proactive and sustainable solution on deepfake detection, that is agnostic to the evolution of generative models, by introducing artificial fingerprints into the models.
Our approach is simple and effective. We first embed artificial fingerprints into training data, then validate a surprising discovery on the transferability of such fingerprints from training data to generative models, which in turn appears in the generated deepfakes. Experiments show that our fingerprinting solution (1) holds for a variety of cutting-edge generative models, (2) leads to a negligible side effect on generation quality, (3) stays robust against image-level and model-level perturbations, (4) stays hard to be detected by adversaries, and (5) converts deepfake detection and attribution into trivial tasks and outperforms the recent state-of-the-art baselines. Our solution closes the responsibility loop between publishing pre-trained generative model inventions and their possible misuses, which makes it independent of the current arms race
Inputs from Hell: Learning Input Distributions for Grammar-Based Test Generation
Get PDFGrammars can serve as producers for structured test inputs that are syntactically correct by construction. A probabilistic grammar assigns probabilities to individual productions, thus controlling the distribution of input elements. Using the grammars as input parsers, we show how to learn input distributions from input samples, allowing to create inputs that are similar to the sample; by inverting the probabilities, we can create inputs that are dissimilar to the sample. This allows for three test generation strategies:
1) “Common inputs” – by learning from common inputs, we can create inputs that are similar to the sample; this is useful for regression testing.
2) “Uncommon inputs” – learning from common inputs and inverting probabilities yields inputs that are strongly dissimilar to the sample; this is useful for completing a test suite with “inputs from hell” that test uncommon features, yet are syntactically valid.
3) “Failure-inducing inputs” – learning from inputs that caused failures in the past gives us inputs that share similar features and thus also have a high chance of triggering bugs; this is useful for testing the completeness of fixes.
Our evaluation on three common input formats (JSON, JavaScript, CSS) shows the effectiveness of these approaches. Results show that “common inputs” reproduced 96% of the methods induced by the samples. In contrast, for almost all subjects (95%), the “uncommon inputs” covered significantly different methods from the samples. Learning from failure-inducing samples reproduced all exceptions (100%) triggered by the failure-inducing samples and discovered new exceptions not found in any of the samples learned from
Slitheen++: Stealth TLS-based Decoy Routing
Get PDFWe present Slitheen++, a decoy routing system that---in contrast to its predecessor Slitheen---is not susceptible to traffic analysis in the upstream channel. Slitheen++ overcomes key challenges such as scheduling for covert connections and technologies to more realistically emulate a real user's behavior, such as crawling or delaying overt communication. We measure Slitheen++ according to metrics that not only show the maximum theoretical throughput of the system, but for the first time, also assess the actual user experience by measuring loading times of websites from ten covert targets. We show that emulating a user increases loading times, yet raises the difficulty for an advanced censor to expose decoy routing as such. For example, crawling raises the median of the loading time for covert setups by 1 second from 7s to 8s
Key Negotiation Downgrade Attacks on Bluetooth and Bluetooth Low Energy
No full textBluetooth (BR/EDR) and Bluetooth Low Energy (BLE) are pervasive wireless technologies specified in the Bluetooth standard. The standard includes key negotiation protocols used to generate long term keys (during pairing) and session keys (during secure connection establishment). In this work, we demonstrate that the key negotiation protocols of Bluetooth and BLE are vulnerable to standard-compliant entropy downgrade attacks. In particular, we show how an attacker can downgrade the entropy of any Bluetooth session key to 1 byte, and of any BLE long term key and session key to 7 bytes. Such low entropy values enable the attacker to brute force Bluetooth long term keys and BLE long term and session keys, and to break all the security guarantees promised by Bluetooth and BLE. As a result of our attacks, an attacker can decrypt all the ciphertext and inject valid ciphertext in any Bluetooth and BLE network.
Our key negotiation downgrade attacks are conducted remotely, do not require access to the victims’ devices and are stealthy to the victims. As the attacks are standard-compliant, they are effective regardless of the usage of the strongest Bluetooth and BLE security modes (including Secure Connections), the Bluetooth version, and the implementation details of the devices used by the victims. We successfully attack 38 Bluetooth devices (32 unique Bluetooth chips) and 19 BLE devices from different vendors, using all the major versions of the Bluetooth standard. Finally, we present effective legacy compliant and non-legacy compliant countermeasures to mitigate our key negotiation downgrade attacks
Constrained Concealment Attacks against Reconstruction-based Anomaly Detectors in Industrial Control Systems
Get PDFRecently, reconstruction-based anomaly detection was proposed as an effective technique to detect attacks in dynamic industrial control networks. Unlike classical network anomaly detectors that observe the network traffic, reconstruction-based detectors operate on the measured sensor data, leveraging physical process models learned a priori.
In this work, we investigate different approaches to evade prior-work reconstruction-based anomaly detectors by manipulating sensor data so that the attack is concealed. We find that replay attacks (commonly assumed to be very strong) show bad performance (i.e., increasing the number of alarms) if the attacker is constrained to manipulate less than 95% of all features in the system, as hidden correlations between the features are not replicated well. To address this, we propose two novel attacks that manipulate a subset of the sensor readings, leveraging learned physical constraints of the system. Our attacks feature two different attacker models: A whitebox attacker, which uses an optimization approach with a detection oracle, and a blackbox attacker, which uses an autoencoder to translate anomalous data into normal data.
We evaluate our implementation on two different datasets from the water distribution domain, showing that the detector's Recall drops from 0.68 to 0.12 by manipulating 4 sensors out of 82 in WADI dataset. In addition, we show that our blackbox attacks are transferable to different detectors: They work against autoencoder-, LSTM-, and CNN-based detectors. Finally, we implement and demonstrate our attacks on a real industrial testbed to demonstrate their feasibility in real-time
Identifying Domains of Applicability of Machine Learning Models for Materials Science
Get PDFAlthough machine learning (ML) models promise to substantially accelerate the discovery of
novel materials, their performance is often still insufficient to draw reliable conclusions.
Improved ML models are therefore actively researched, but their design is currently guided
mainly by monitoring the average model test error. This can render different models indistinguishable
although their performance differs substantially across materials, or it can make
a model appear generally insufficient while it actually works well in specific sub-domains.
Here, we present a method, based on subgroup discovery, for detecting domains of applicability
(DA) of models within a materials class. The utility of this approach is demonstrated
by analyzing three state-of-the-art ML models for predicting the formation energy of
transparent conducting oxides. We find that, despite having a mutually indistinguishable and
unsatisfactory average error, the models have DAs with distinctive features and notably
improved performance
Better Code, Better Sharing: On the Need of Analyzing Jupyter Notebooks
Get PDFBy bringing together code, text, and examples, Jupyter notebooks have become one of the most popular means to produce scientific results in a productive and reproducible way. As many of the notebook authors are experts in their scientific fields, but laymen with respect to software engineering, one may ask questions on the quality of notebooks and their code. In a preliminary study, we experimentally demonstrate that Jupyter notebooks are inundated with poor quality code, e.g., not respecting recommended coding practices, or containing unused variables and deprecated functions. Considering the education nature of Jupyter notebooks, these poor coding practices as well as the lacks of quality control might be propagated into the next generation of developers. Hence, we argue that there is a strong need to programmatically analyze Jupyter notebooks, calling on our community to pay more attention to the reliability of Jupyter notebooks