WOOT '18 Workshop Program

The history of security includes a long series of arms races, where a new technology emerges and is subsequently developed and exploited by both defenders and attackers. Over the past few years, "Artificial Intelligence" has re-emerged as a potentially transformative technology, and deep learning in particular has produced a barrage of amazing results. We are in the very early stages of understanding the potential of this technology in security, but more worryingly, seeing how it may be exploited by malicious individuals and powerful organizations. In this talk, I'll look at what lessons might be learned from previous security arms races, consider how asymmetries in AI may be exploited by attackers and defenders, touch on some recent work in adversarial machine learning, and hopefully help progress-loving Luddites figure out how to survive in a world overrun by AI doppelgängers, GAN gangs, and gibbon-impersonating pandas.

Deep neural networks (DNNs) are vulnerable to adversarial examples—maliciously crafted inputs that cause DNNs to make incorrect predictions. Recent work has shown that these attacks generalize to the physical domain, to create perturbations on physical objects that fool image classifiers under a variety of real-world conditions. Such attacks pose a risk to deep learning models used in safety-critical cyber-physical systems. In this work, we extend physical attacks to more challenging object detection models, a broader class of deep learning algorithms widely used to detect and label multiple objects within a scene. Improving upon a previous physical attack on image classifiers, we create perturbed physical objects that are either ignored or mislabeled by object detection models. We implement a Disappearance Attack, in which we cause a Stop sign to “disappear” according to the detector—either by covering the sign with an adversarial Stop sign poster, or by adding adversarial stickers onto the sign. In a video recorded in a controlled lab environment, the state-of-the-art YOLO v2 detector failed to recognize these adversarial Stop signs in over 85% of the video frames. In an outdoor experiment, YOLO was fooled by the poster and sticker attacks in 72.5% and 63.5% of the video frames respectively. We also use Faster R-CNN, a different object detection model, to demonstrate the transferability of our adversarial perturbations. The created poster perturbation is able to fool Faster R-CNN in 85.9% of the video frames in a controlled lab environment, and 40.2% of the video frames in an outdoor environment. Finally, we present preliminary results with a new Creation Attack, wherein innocuous physical stickers fool a model into detecting nonexistent objects.

The current generation of low- and medium interaction honeypots uses off-the-shelf libraries to provide the transport layer. We show that this architecture is fatally flawed because the protocols are implemented subtly differently from the systems being impersonated. We present a generic technique for systematically fingerprinting low- and medium interaction honeypots at Internet scale with just one packet and an ERR (Equal Error Rate) of 0.0183. We conduct Internet-wide scans and identify 7,605 honeypot instances across nine different honeypot implementations for the most important network protocols SSH, Telnet, and HTTP. For SSH honeypots we also determined their patch level and find that they are poorly maintained -- 27% of the honeypots have not been updated within the last 31 months and only 39% incorporate improvements from 7 months ago. We believe our findings to be a `class break' in that trivial patches cannot address the issue.

Shrew attacks or pulsing attacks are low-bandwidth network-level/layer-3 denial-of-service attacks. They target TCP connections by selectively inducing packet loss to affect latency and throughput. We combine the recently presented DNS CNAME-chaining attack with temporal lensing, a variant of pulsing attacks, to create a new, harder to block attack. For an attack, thousands of DNS resolvers have to be coordinated. We devise an optimization problem to find the perfect attack and solve it by using a genetic algorithm. The results show pulses created with our attack are 14 times higher than the attacker's average bandwidth. Finally, we present countermeasures applicable to pulsing and CNAME-chaining, which also apply to this attack.

We show how to efficiently simulate cryptographic primitives during symbolic execution. This allows analysis of security protocol implementations, and revealed several flaws in implementations of WPA2's 4-way handshake.

Traditional symbolic execution engines cannot handle cryptographic primitives, because analyzing them results in complex symbolic expressions that cannot be handled by the SMT solver. We prevent this by simulating their behaviour under the Dolev-Yao model. This enables efficient symbolic execution of security protocols implementations, making it possible to detect common programming mistakes in them. We also show how to detect misuse of cryptographic primitives. That is, we can detect trivial timing side-channels, and we can identify decryption oracles where unauthenticated decrypted data influences the program's behaviour. We apply our technique on three client-side implementations of WPA2's 4-way handshake. This uncovered timing side-channels when verifying authentication tags, a denial-of-service attack, a stack-based buffer overflow, and also revealed a non-trivial decryption oracle. We confirmed all vulnerabilities in practice, and discuss them in detail.

In 2014, the European Commission released the eIDAS regulation to target the compatibility of cross-country electronic services within the European Union. eIDAS (electronic IDentification, Authentication, and Trust Ser- vices) defines implementation standards and technologies for electronic signatures, digital certificates, Single Sign-On (SSO), and trust services. It is based on well-established standards, such as SAML, to achieve high security and compatibility between EU countries. In this paper, we present the first security study of authentication schemes used in eID services. Our security analysis shows that 7 of the 15 European eID services were vulnerable to XML-based attacks which enabled efficient Denial-of-Service (DoS) and Server Side Request Forgery (SSRF) attacks. On 5 of the 15 eID services, we were even able to exfiltrate locally stored files and send these files to an arbitrary domain. To support the developers and security teams of eID services, we implemented a Burp Suite extension to execute fully-automated or semi-automated tests. Additionally, we summarize best practices related to eID-based authentication and SSO in general.

Since its creation, SSL/TLS has been the go-to solution for securing unencrypted web protocols - most commonly HTTP. The design of SSL/TLS, however, merely provides data stream encryption and authentication properties which often leads to the incorrect conclusion that by simply wrapping an unencrypted HTTP connection to a server with SSL/TLS, user privacy and web application behaviour integrity are guaranteed. Such type of information leak is unique in the sense that while certain web security vulnerabilities such as SQL injections have been well researched and thus there are known design patterns to avoid and penetration testing tools based on detecting known-to-be insecure design patterns, the state of research for the types of information leaks described in this paper still lags behind. In this paper, we discuss three design patterns that often result in side-channel information leaks along with three real-world websites which posses these vulnerabilities. Based on these three vulnerable design patterns we present a set of tools for detecting these types of side-channel information leaks given a training set of captured encrypted network traffic sessions.