WebAssembly, the open standard for binary code, is quickly gaining adoption on the web and beyond. As the binaries are often written in low-level languages, like C and C++, they are riddled with the same bugs as their traditional counterparts. Minimal tooling to uncover these bugs on WebAssembly binaries exists. In this paper we present WAFL, a fuzzer for WebAssembly binaries. WAFL adds a set of patches to the WAVM WebAssembly runtime to generate coverage data for the popular AFL++ fuzzer. Thanks to the underlying ahead-of-time (AOT) compiling WAVM, WAFL is already very performant. WAFL adds lightweight VM snapshots. By replacing forks, traditionally used in AFL++ harnesses, with WAFL’s snapshots, WAFL harnesses can even outperform native harnesses with compile-time instrumentation in raw fuzzing performance. To the best of our knowledge, WAFL is the first coverage-guided fuzzer for binary-only WebAssembly, without the need for source.
Google recently introduced a secure chip called Titan M in its Pixel smartphones, enabling the implementation of a Trusted Execution Environment (TEE) in Tamper Resistant Hardware. TEEs have been proven effective in reducing the attack surface exposed by smartphones, by protecting specific security-sensitive operations. However, studies have shown that TEE code and execution can also be targeted and exploited by attackers, therefore, studying their security lays the basis of the trust we have in their features. In this paper, we provide the first security analysis of Titan M. First, we reverse engineer the firmware and we review the open source code in the Android OS that is responsible for the communication with the chip. By exploiting a known vulnerability, we then dynamically examine the memory layout and the internals of the chip. Finally, leveraging the acquired knowledge, we design and implement a structure-aware black-box fuzzer. Using our fuzzer, we rediscover several known vulnerabilities after a few seconds of testing, proving the effectiveness of our solution. In addition, we identify and report a new vulnerability in the latest version of the firmware.
The ongoing public interest in blockchains and smart contracts has brought a rise to a magnitude of different blockchain implementations. The rate at which new concepts are envisioned and implemented makes it hard to vet their impact on security. Especially smart contract platforms, executing untrusted code, are very complex by design. Still, people put their trust and money into chains that may lack proper testing. A behavior deviation for edge cases of single op-codes is a critical bug class in this brave new world. It can be abused for Denial of Service against the blockchain, chain splits, double-spending, or direct attacks on applications operating on the blockchain. In this paper, we propose an automated methodology to uncover such differences. Through coverage-guided and state-guided fuzzing, we explore smart contract virtual machine behavior against multiple VMs in parallel. We develop NeoDiff, the first framework for feedback-guided differential fuzzing of smart contract VMs. We discuss real, monetary consequences our tool prevents. NeoDiff can be ported to new smart contract platforms with ease. Apart from fuzzing Ethereum VMs, NeoDiff found a range of critical differentials in VMs for the Neo blockchain. Moreover, through a higher-layer semantics mutator, we uncovered semantic discrepancies between Neo smart contracts written in Python when executed on the blockchain vs. classic CPython. Along the way, NeoDiff uncovered memory corruptions in the C# Neo VM.
Adversarial attacks to Machine Learning (ML) models became such a concern that tech companies (Microsoft and CUJO AI’s Vulnerability Research Lab) decided to launch contests to better understand their impact on practice. During the contest’s first edition (2019), participating teams were challenged to bypass three ML models in a white box manner. Our team bypassed all the three of them and reported interesting insights about models’ weaknesses. In the second edition (2020), the challenge evolved to an attack-and-defense model: the teams should either propose defensive models and attack other teams’ models in a black box manner. Despite the difficulty increase, our team was able to bypass all models again. In this paper, we describe our insights for this year’s contest regarding on attacking models, as well defending them from adversarial attacks. In particular, we show how frequency-based models (e.g., TF-IDF) are vulnerable to the addition of dead function imports, and how models based on raw bytes are vulnerable to payload-embedding obfuscation (e.g., XOR and base64 encoding).
Machine learning based solutions have been very helpful in solving problems that deal with immense amounts of data, such as malware detection and classification. However, deep neural networks have been found to be vulnerable to adversarial examples, or inputs that have been purposefully perturbed to result in an incorrect label. Researchers have shown that this vulnerability can be exploited to create evasive malware samples. However, many proposed attacks do not generate an executable and instead generate a feature vector. To fully understand the impact of adversarial examples on malware detection, we review practical attacks against malware classifiers that generate executable adversarial malware examples. We also discuss current challenges in this area of research, as well as suggestions for improvement and future research directions.
