USENIX Security '20 Technical Sessions

The IEEE 802.11 WPA2 protocol is widely used across the globe to protect network connections. The protocol, which is specified on more than three-thousand pages and has received various patches over the years, is extremely complex and therefore hard to analyze. In particular, it involves various mechanisms that interact with each other in subtle ways, which offers little hope for modular reasoning. Perhaps because of this, there exists no formal or cryptographic argument that shows that the patches to the core protocol indeed prevent the corresponding attacks, such as, e.g., the notorious KRACK attacks from 2017.

In this work, we address this situation and present an extensive formal analysis of the WPA2 protocol design. Our model is the first that is detailed enough to detect the KRACK attacks; it includes mechanisms such as the four-way handshake, the group-key handshake, WNM sleep mode, the data-confidentiality protocol, and their complex interactions.

Our analysis provides the first security argument, in any formalism, that the patched WPA2 protocol meets its claimed security guarantees in the face of complex modern attacks.

To defeat security threats such as man-in-the-middle (MITM) attacks, Bluetooth Low Energy (BLE) 4.2 and 5.x introduced a Secure Connections Only (SCO) mode, under which a BLE device can only accept secure pairing such as Passkey Entry and Numeric Comparison from an initiator, e.g., an Android mobile. However, the BLE specification does not require the SCO mode for the initiator, and does not specify how the BLE programming framework should implement this mode. In this paper we show that the BLE programming framework of the initiator must properly handle SCO initiation, status management, error handling, and bond management; otherwise severe flaws can be exploited to perform downgrade attacks, forcing the BLE pairing protocols to run in an insecure mode without user's awareness. To validate our findings, we have tested 18 popular BLE commercial products with 5 Android phones. Our experimental results proved that MITM attacks (caused by downgrading) are possible to all these products. More importantly, due to such system flaws from the BLE programming framework, all BLE apps in Android are subject to our downgrade attacks. To defend against our attacks, we have built a prototype for the SCO mode on Android 8 atop Android Open Source Project (AOSP). Finally, in addition to Android, we also find all major OSes including iOS, macOS, Windows, and Linux do not support the SCO mode properly. We have reported the identified BLE pairing vulnerabilities to Bluetooth Special Interest Group, Google, Apple, Texas Instruments, and Microsoft.

Voice over LTE (VoLTE) is a packet-based telephony service seamlessly integrated into the Long Term Evolution (LTE) standard and deployed by most telecommunication providers in practice. Due to this widespread use, successful attacks against VoLTE can affect a large number of users worldwide. In this work, we introduce ReVoLTE, an attack that exploits an LTE implementation flaw to recover the contents of an encrypted VoLTE call, hence enabling an adversary to eavesdrop on phone calls. ReVoLTE makes use of a predictable keystream reuse on the radio layer that allows an adversary to decrypt a recorded call with minimal resources. Through a series of preliminary as well as real-world experiments, we successfully demonstrate the feasibility of ReVoLTE and analyze various factors that critically influence our attack in commercial networks. For mitigating the ReVoLTE attack, we propose and discuss short- and long-term countermeasures deployable by providers and equipment vendors.

The Android operating system is the world's dominant mobile computing platform. To defend against malicious applications and external attack, Android relies upon a complex combination of discretionary and mandatory access control mechanisms, including Linux capabilities, to maintain least privilege. To understand the impact and interaction between these layers, we created a framework called BigMAC that combines and instantiates all layers of the policy together in a fine grained graph supporting millions of edges. Our model filters out paths and types not in use on actual systems that policy analysis alone would consider. Unlike previous work which requires a rooted device, using only static firmware and Android domain knowledge, we are able to extract and recreate the security state of a running system, achieving a process credential recovery at best 74.7% and a filesystem DAC and MAC accuracy of over 98%. Using BigMAC, we develop attack queries to discover sets of objects that can be influenced by untrusted applications and external peripherals. Our evaluation against Samsung S8+ and LG G7 firmwares reveals multiple policy concerns, including untrusted apps on LG being able to communicate with a kernel monitoring service, Samsung S8+ allowing IPC from untrusted apps to some root processes, at least 24 processes with the CAP_SYS_ADMIN capability, and system_server with the capability to load kernel modules. We have reported our findings to the corresponding vendors and release BigMAC for the community.

Android native system services provide essential supports and fundamental functionalities for user apps. Finding vulnerabilities in them is crucial for Android security. Fuzzing is one of the most popular vulnerability discovery solutions, yet faces several challenges when applied to Android native system services. First, such services are invoked via a special interprocess communication (IPC) mechanism, namely binder, via service-specific interfaces. Thus, the fuzzer has to recognize all interfaces and generate interface-specific test cases automatically. Second, effective test cases should satisfy the interface model of each interface. Third, the test cases should also satisfy the semantic requirements, including variable dependencies and interface dependencies.

In this paper, we propose an automated generation-based fuzzing solution FANS to find vulnerabilities in Android native system services. It first collects all interfaces in target services and uncovers deep nested multi-level interfaces to test. Then, it automatically extracts interface models, including feasible transaction code, variable names and types in the transaction data, from the abstract syntax tree (AST) of target interfaces. Further, it infers variable dependencies in transactions via the variable name and type knowledge, and infers interface dependencies via the generation and use relationship. Finally, it employs the interface models and dependency knowledge to generate sequences of transactions, which have valid formats and semantics, to test interfaces of target services. We implemented a prototype of FANS from scratch and evaluated it on six smartphones equipped with a recent version of Android, i.e., android-9.0.0_r46 , and found 30 unique vulnerabilities deduplicated from thousands of crashes, of which 20 have been confirmed by Google. Surprisingly, we also discovered 138 unique Java exceptions during fuzzing.

Nowadays, virtually all mobile apps rely on communicating with a network backend. Given the sensitive nature of the data exchanged between apps and their backends, securing these network communications is of growing importance. In recent years, Google has developed a number of security mechanisms for Android apps, ranging from multiple KeyStores to the recent introduction of the new Network Security Policy, an XML-based configuration file that allows apps to define their network security posture. In this paper, we perform the first comprehensive study on these new network defense mechanisms. In particular, we present them in detail, we discuss the attacks they are defending from, and the relevant threat models. We then discuss the first large-scale analysis on this aspect. During June and July 2019, we crawled 125,419 applications and we found how only 16,332 apps adopt this new security feature. We then focus on these apps, and we uncover how developers adopt weak and potentially vulnerable network security configurations. We note that, in November 2019, Google then made the default policy stricter, which would help the adoption. We thus opted to re-crawl the same dataset (from April to June 2020) and we repeated the experiments: while more apps do adopt this new security mechanism, a significant portion of them still do not take fully advantage of it (e.g., by allowing usage of insecure protocols). We then set out to explore the root cause of these weaknesses (i.e., the why). Our analysis showed that app developers often copy-paste vulnerable policies from popular developer websites (e.g., StackOverflow). We also found that several popular ad libraries require apps to weaken their security policy, the key problem lying in the vast complexity of the ad ecosystem. As a last contribution, we propose a new extension of the Network Security Policy, so to allow app developers to embed problematic ad libraries without the need to weaken the security of their entire app.

An exciting insight of recent networking research has been that path-aware networking architectures are able to fundamentally solve many of the security issues of today’s Internet, while increasing overall efficiency and giving control over path selection to end hosts. In this paper, we consider three important issues related to this new networking paradigm: First, network operators still need to be able to impose their own policies to rule out uneconomical paths and to enforce these decisions on the data plane. Second, end hosts should be able to verify that their forwarding decisions are actually followed by the network. Finally, both intermediate routers and recipients should be able to authenticate the source of packets. These properties have been considered by previous work, but there is no existing system that achieves both strong security guarantees and high efficiency.

We propose EPIC, a family of data-plane protocols that provide increasingly strong security properties, addressing all three described requirements. The EPIC protocols have significantly lower communication overhead than comparable systems: for realistic path lengths, the overhead is 3–5 times smaller compared to the state-of-the-art systems OPT and ICING. Our prototype implementation is able to saturate a 40 Gbps link even on commodity hardware due to the use of only few highly efficient symmetric cryptographic operations in the forwarding process. Thus, by ensuring that every packet is checked at every hop, we make an important step towards an efficient and secure future Internet.

In today's DNS infrastructure, DNS forwarders are devices standing in between DNS clients and recursive resolvers. The devices often serve as ingress servers for DNS clients, and instead of resolving queries, they pass the DNS requests to other servers. Because of the advantages and several use cases, DNS forwarders are widely deployed and queried by Internet users. However, studies have shown that DNS forwarders can be more vulnerable devices in the DNS infrastructure.

In this paper, we present a cache poisoning attack targeting DNS forwarders. Through this attack, attackers can inject rogue records of arbitrary victim domain names using a controlled domain, and circumvent widely-deployed cache poisoning defences. By performing tests on popular home router models and DNS software, we find several vulnerable implementations, including those of large vendors (e.g., D-Link, Linksys, dnsmasq and MS DNS). Further, through a nationwide measurement, we estimate the population of Chinese mobile clients which are using vulnerable DNS forwarders. We have been reporting the issue to the affected vendors, and so far have received positive feedback from three of them. Our work further demonstrates that DNS forwarders can be a soft spot in the DNS infrastructure, and calls for attention as well as implementation guidelines from the community.

The DNS-based Authentication of Named Entities (DANE) standard allows clients and servers to establish a TLS connection without relying on trusted third parties like CAs by publishing TLSA records. DANE uses the Domain Name System Security Extensions (DNSSEC) PKI to achieve integrity and authenticity. However, DANE can only work correctly if each principal in its PKI properly performs its duty: through their DNSSEC-aware DNS servers, DANE servers (e.g., SMTP servers) must publish their TLSA records, which are consistent with their certificates. Similarly, DANE clients (e.g., SMTP clients) must verify the DANE servers’ TLSA records, which are also used to validate the fetched certificates.

DANE is rapidly gaining popularity in the email ecosystem, to help improve transport security between mail servers. Yet its security benefits hinge on deploying DANE correctly. In this paper we perform a large-scale, longitudinal, and comprehensive measurement study on how well the DANE standard and its relevant protocols are deployed and managed. We collect data for all second-level domains under the .com, .net, .org, .nl, and .se TLDs over a period of 24 months to analyze server-side deployment and management. To analyse the client-side deployment and management, we investigate 29 popular email service providers, and four popular MTA and ten DNS software programs.

Our study reveals pervasive mismanagement in the DANE ecosystem. For instance, we found that 36% of TLSA records cannot be validated due to missing or incorrect DNSSEC records, and 14.17% of them are inconsistent with their certificates. We also found that only four email service providers support DANE for both outgoing and incoming emails, but two of them have drawbacks of not checking the Certificate Usage in TLSA records. On the bright side, the administrators of email servers can leverage open source MTA and DNS programs to support DANE correctly.

Autonomous vehicles are becoming increasingly popular, but their reliance on computer systems to sense and operate in the physical world introduces new security risks. In this paper, we show that the location privacy of an autonomous vehicle may be compromised by software side-channel attacks if localization software shares a hardware platform with an attack program. In particular, we demonstrate that a cache side-channel attack can be used to infer the route or the location of a vehicle that runs the adaptive Monte-Carlo localization (AMCL) algorithm. The main contributions of the paper are as follows. First, we show that adaptive behaviors of perception and control algorithms may introduce new side-channel vulnerabilities that reveal the physical properties of a vehicle or its environment. Second, we introduce statistical learning models that infer the AMCL algorithm's state from cache access patterns and predict the route or the location of a vehicle from the trace of the AMCL state. Third, we implement and demonstrate the attack on a realistic software stack using real-world sensor data recorded on city roads. Our findings suggest that autonomous driving software needs strong timing-channel protection for location privacy.

Autonomous Vehicles (AVs), including aerial, sea, and ground vehicles, assess their environment with a variety of sensors and actuators that allow them to perform specific tasks such as navigating a route, hovering, or avoiding collisions. So far, AVs tend to trust the information provided by their sensors to make navigation decisions without data validation or verification, and therefore, attackers can exploit these limitations by feeding erroneous sensor data with the intention of disrupting or taking control of the system. In this paper we introduce SAVIOR: an architecture for securing autonomous vehicles with robust physical invariants. We implement and validate our proposal on two popular open-source controllers for aerial and ground vehicles, and demonstrate its effectiveness.

For high-level Autonomous Vehicles (AV), localization is highly security and safety critical. One direct threat to it is GPS spoofing, but fortunately, AV systems today predominantly use Multi-Sensor Fusion (MSF) algorithms that are generally believed to have the potential to practically defeat GPS spoofing. However, no prior work has studied whether today's MSF algorithms are indeed sufficiently secure under GPS spoofing, especially in AV settings. In this work, we perform the first study to fill this critical gap. As the first study, we focus on a production-grade MSF with both design and implementation level representativeness, and identify two AV-specific attack goals, off-road and wrong-way attacks.

To systematically understand the security property, we first analyze the upper-bound attack effectiveness, and discover a take-over effect that can fundamentally defeat the MSF design principle. We perform a cause analysis and find that such vulnerability only appears dynamically and non-deterministically. Leveraging this insight, we design FusionRipper, a novel and general attack that opportunistically captures and exploits take-over vulnerabilities. We evaluate it on 6 real-world sensor traces, and find that FusionRipper can achieve at least 97% and 91.3% success rates in all traces for off-road and wrong-way attacks respectively. We also find that it is highly robust to practical factors such as spoofing inaccuracies. To improve the practicality, we further design an offline method that can effectively identify attack parameters with over 80% average success rates for both attack goals, with the cost of at most half a day. We also discuss promising defense directions.

IoT clouds facilitate the communication between IoT devices and users, and authorize users’ access to their devices. In this paradigm, an IoT device is usually managed under a particular IoT cloud designated by the device vendor, e.g., Philips bulbs are managed under Philips Hue cloud. Today’s mainstream IoT clouds also support device access delegation across different vendors (e.g., Philips Hue, LIFX, etc.) and cloud providers (e.g., Google, IFTTT, etc.): for example, Philips Hue and SmartThings clouds support to delegate device access to another cloud such as Google Home, so a user can manage multiple devices from different vendors all through Google Home. Serving this purpose are the IoT delegation mechanisms developed and utilized by IoT clouds, which we found are heterogeneous and ad-hoc in the wild, in the absence of a standardized delegation protocol suited for IoT environments. In this paper, we report the first systematic study on real-world IoT access delegation, based upon a semi-automatic verification tool we developed. Our study brought to light the pervasiveness of security risks in these delegation mechanisms, allowing the adversary (e.g., Airbnb tenants, former employees) to gain unauthorized access to the victim’s devices (e.g., smart locks) or impersonate the devices to trigger other devices. We confirmed the presence of critical security flaws in these mechanisms through end-to-end exploits on them, and further conducted a measurement study. Our research demonstrates the serious consequences of these exploits and the security implications of the practice today for building these mechanisms. We reported our findings to related parties, which acknowledged the problems. We further propose principles for developing more secure cross-cloud IoT delegation services, before a standardized solution can be widely deployed.

Microcontroller-based embedded systems are increasingly used for applications that can have serious and immediate consequences if compromised—including automobile control systems, smart locks, drones, and implantable medical devices. Due to resource and execution-time constraints, C is the primary language used for programming these devices. Unfortunately, C is neither type-safe nor memory-safe, and control-flow hijacking remains a prevalent threat.

This paper presents Silhouette: a compiler-based defense that efficiently guarantees the integrity of return addresses, significantly reducing the attack surface for control-flow hijacking. Silhouette combines an incorruptible shadow stack for return addresses with checks on forward control flow and memory protection to ensure that all functions return to the correct dynamic caller. To protect its shadow stack, Silhouette uses store hardening, an efficient intra-address space isolation technique targeting various ARM architectures that leverages special store instructions found on ARM processors.

We implemented Silhouette for the ARMv7-M architecture, but our techniques are applicable to other common embedded ARM architectures. Our evaluation shows that Silhouette incurs a geometric mean of 1.3% and 3.4% performance overhead on two benchmark suites. Furthermore, we prototyped Silhouette-Invert, an alternative implementation of Silhouette, which incurs just 0.3% and 1.9% performance overhead, at the cost of a minor hardware change.

Counterfeit integrated circuits are responsible for billions of dollars in losses to the semiconductor industry each year, and jeopardize the reliability of critical systems that unwittingly rely on them. Counterfeit parts, which are primarily recycled, test rejects, or legitimate but regraded, have to date been found in a number of systems, including critical defense systems. In this work, we present COUNTERFOIL – an anti-counterfeiting system based on enrolling and authenticating intrinsic features of the molded packages that enclose a majority of semiconductor chips sold on the market. Our system relies on computer-readable labels, inexpensive cameras, imaging processing using OpenCV, and digital signatures, to enroll and verify chip packages. We demonstrate our approach on a dataset from over 100 chips. We show that our technique is effective and reliable for verifying provenance under a variety of settings, and evaluate the robustness of the package features by using different imaging platforms, and by wearing the chips with silicon carbide polishing grit in a rock tumbler. We show that, even if an adversary steals the exact mold used to produce an enrolled chip package, he will have limited success in being able to counterfeit the chip.

Since 2016, with a strong push from the Government of India, smartphone-based payment apps have become mainstream, with over $50 billion transacted through these apps in 2018. Many of these apps use a common infrastructure introduced by the Indian government, called the Unified Payments Interface (UPI), but there has been no security analysis of this critical piece of infrastructure that supports money transfers. This paper uses a principled methodology to do a detailed security analysis of the UPI protocol by reverse-engineering the design of this protocol through seven popular UPI apps. We discover previously-unreported multi-factor authentication design-level flaws in the UPI 1.0 specification that can lead to significant attacks when combined with an installed attacker-controlled application. In an extreme version of the attack, the flaws could allow a victim's bank account to be linked and emptied, even if a victim had never used a UPI app. The potential attacks were scalable and could be done remotely. We discuss our methodology and detail how we overcame challenges in reverse-engineering this unpublished application layer protocol, including that all UPI apps undergo a rigorous security review in India and are designed to resist analysis. The work resulted in several CVEs, and a key attack vector that we reported was later addressed in UPI 2.0.

In the 2018 midterm elections, West Virginia became the first state in the U.S. to allow select voters to cast their ballot on a mobile phone via a proprietary app called “Voatz.” Although there is no public formal description of Voatz’s security model, the company claims that election security and integrity are maintained through the use of a permissioned blockchain, biometrics, a mixnet, and hardware-backed key storage modules on the user’s device. In this work, we present the first public security analysis of Voatz, based on a reverse engineering of their Android application and the minimal available documentation of the system. We performed a cleanroom reimplementation of Voatz’s server and present an analysis of the election process as visible from the app itself.

We find that Voatz has vulnerabilities that allow different kinds of adversaries to alter, stop, or expose a user’s vote, including a sidechannel attack in which a completely passive network adversary can potentially recover a user’s secret ballot. We additionally find that Voatz has a number of privacy issues stemming from their use of third party services for crucial app functionality. Our findings serve as a concrete illustration of the common wisdom against Internet voting, and of the importance of transparency to the legitimacy of elections. As a result of our work, West Virginia and one county in Washington has already aborted their use of Voatz in the 2020 primaries.

Card-not-present credit card fraud costs businesses billions of dollars a year. In this paper, we present Boxer, a mobile SDK and server that enables apps to combat card-not-present fraud by scanning cards and verifying that they are genuine. Boxer analyzes the images from these scans, looking for tell-tale signs of attacks, and introduces a novel abstraction on top of modern security hardware for complementary protection.

Currently, 323 apps have integrated Boxer, and tens of them have deployed it to production, including some large, popular, and international apps, resulting in Boxer scanning over 10 million real cards already. Our evaluation of Boxer from one of these deployments shows ten cases of real attacks that our novel hardware-based abstraction detects. Additionally, from the same deployment, without letting in any fraud, Boxer’s card scanning recovers 89% of the good users whom the app would have blocked. In another evaluation of Boxer, we run our image analysis models against images from real users and show an accuracy of 96% and 100% on the two models that we use.

Side-channel attacks exploiting (EC)DSA nonce leakage easily lead to full key recovery. Although (EC)DSA implementations have already been hardened against side-channel leakage using the constant-time paradigm, the long-standing cat-and-mouse-game of attacks and patches continues. In particular, current code review is prone to miss less obvious side channels hidden deeply in the call stack. To solve this problem, a systematic study of nonce leakage is necessary. We present a systematic analysis of nonce leakage in cryptographic implementations. In particular, we expand DATA, an open-source side-channel analysis framework, to detect nonce leakage. Our analysis identified multiple unknown nonce leakage vulnerabilities across all essential computation steps involving nonces. Among others, we uncover inherent problems in Bignumber implementations that break claimed constant-time guarantees of (EC)DSA implementations if secrets are close to a word boundary. We found that lazy resizing of Bignumbers in OpenSSL and LibreSSL yields a highly accurate and easily exploitable side channel, which has been acknowledged with two CVEs. Surprisingly, we also found a tiny but expressive leakage in the constant-time scalar multiplication of OpenSSL and BoringSSL. Moreover, in the process of reporting and patching, we identified newly introduced leakage with the support of our tool, thus preventing another attack-patch cycle. We open-source our tool, together with an intuitive graphical user interface we developed.

The security of FPGAs is a crucial topic, as any vulnerability within the hardware can have severe consequences, if they are used in a secure design. Since FPGA designs are encoded in a bitstream, securing the bitstream is of the utmost importance. Adversaries have many motivations to recover and manipulate the bitstream, including design cloning, IP theft, manipulation of the design, or design subversions e.g., through hardware Trojans. Given that FPGAs are often part of cyber-physical systems e.g., in aviation, medical, or industrial devices, this can even lead to physical harm. Consequently, vendors have introduced bitstream encryption, offering authenticity and confidentiality. Even though attacks against bitstream encryption have been proposed in the past, e.g., side-channel analysis and probing, these attacks require sophisticated equipment and considerable technical expertise.

In this paper, we introduce novel low-cost attacks against the Xilinx 7-Series (and Virtex-6) bitstream encryption, resulting in the total loss of authenticity and confidentiality. We exploit a design flaw which piecewise leaks the decrypted bitstream. In the attack, the FPGA is used as a decryption oracle, while only access to a configuration interface is needed. The attack does not require any sophisticated tools and, depending on the target system, can potentially be launched remotely. In addition to the attacks, we discuss several countermeasures.

The SHA-1 hash function was designed in 1995 and has been widely used during two decades. A theoretical collision attack was first proposed in 2004 [WYY05], but due to its high complexity it was only implemented in practice in 2017, using a large GPU cluster [SBK+17]. More recently, an almost practical chosen-prefix collision attack against SHA-1 has been proposed [LP19]. This more powerful attack allows to build colliding messages with two arbitrary prefixes, which is much more threatening for real protocols. In this paper, we report the first practical implementation of this attack, and its impact on real-world security with a PGP/GnuPG impersonation attack. We managed to significantly reduce the complexity of collision attacks against SHA-1: on an Nvidia GTX 970, identical-prefix collisions can now be computed with a complexity (expressed in terms of SHA-1 equivalents on this GPU) of 2^61.2 rather than 2^64.7, and chosen-prefix collisions with a complexity of 2^63.4 rather than 2^67.1. When renting cheap GPUs, this translates to a cost of US$ 11k for a collision, and US$ 45k for a chosen-prefix collision, within the means of academic researchers. Our actual attack required two months of computations using 900 Nvidia GTX 1060 GPUs. Therefore, the same attacks that have been practical on MD5 since 2009 are now practical on SHA-1. In particular, chosen-prefix collisions can break signature schemes and handshake security in secure channel protocols (TLS, SSH), if generated extremely quickly. We strongly advise to remove SHA-1 from those type of applications as soon as possible. We exemplify our cryptanalysis by creating a pair of PGP/GnuPG keys with different identities, but colliding SHA-1 certificates. A SHA-1 certification of the first key can therefore be transferred to the second key, leading to an impersonation attack. This proves that SHA-1 signatures now offer virtually no security in practice. The legacy branch of GnuPG still uses SHA-1 by default for identity certifications, but after notifying the authors, the modern branch now rejects SHA-1 signatures (the issue is tracked as CVE-2019-14855).

Verifiable outsourcing systems offload a large computation to a remote server, but require that the remote server provide a succinct proof, called a SNARK, that proves that the server carried out the computation correctly. Real-world applications of this approach can be found in several blockchain systems that employ verifiable outsourcing to process a large number of transactions off-chain. This reduces the on-chain work to simply verifying a succinct proof that transaction processing was done correctly. In practice, verifiable outsourcing of state updates is done by updating the leaves of a Merkle tree, recomputing the resulting Merkle root, and proving using a SNARK that the state update was done correctly.

In this work, we use a combination of existing and novel techniques to implement an RSA accumulator inside of a SNARK, and use it as a replacement for a Merkle tree. We specifically optimize the accumulator for compatibility with SNARKs. Our experiments show that the resulting system can dramatically reduce costs compared to existing approaches that use Merkle trees for committing to the current state. These results apply broadly to any system that needs to offload batches of state updates to a remote untrusted server.

The k-Nearest Neighbor Search (k-NNS) is the backbone of several cloud-based services such as recommender systems, face recognition, and database search on text and images. In these services, the client sends the query to the cloud server and receives the response in which case the query and response are revealed to the service provider. Such data disclosures are unacceptable in several scenarios due to the sensitivity of data and/or privacy laws.

In this paper, we introduce SANNS, a system for secure k-NNS that keeps client's query and the search result confidential. SANNS comprises two protocols: an optimized linear scan and a protocol based on a novel sublinear time clustering-based algorithm. We prove the security of both protocols in the standard semi-honest model. The protocols are built upon several state-of-the-art cryptographic primitives such as lattice-based additively homomorphic encryption, distributed oblivious RAM, and garbled circuits. We provide several contributions to each of these primitives which are applicable to other secure computing tasks. Both of our protocols rely on a new circuit for the approximate top-k selection from n numbers that is built from O(n + k²) comparators.

We have implemented our proposed system and performed extensive experimental results on four datasets in two different computation environments, demonstrating more than 18 — 31 × faster response time compared to optimally implemented protocols from the prior work. Moreover, SANNS is the first work that scales to the database of 10 million entries, pushing the limit by more than two orders of magnitude.

In this work, we consider distributed private learning. For this purpose, companies collect statistics about telemetry, usage and frequent settings from their users without disclosing individual values. We focus on rank-based statistics, specifically, the median which is more robust to outliers than the mean.

Local differential privacy, where each user shares locally perturbed data with an untrusted server, is often used in private learning but does not provide the same accuracy as the central model, where noise is applied only once by a trusted server. Existing solutions to compute the differentially private median provide good accuracy only for large amounts of users (local model), by using a trusted third party (central model), or for a very small data universe (secure multi-party computation).

We present a multi-party computation to efficiently compute the exponential mechanism for the median, which also supports, e.g., general rank-based statistics (e.g., p^th-percentile, interquartile range) and convex optimizations for machine learning. Our approach is efficient (practical running time), scaleable (sublinear in the data universe size) and accurate, i.e., the absolute error is smaller than comparable methods and is independent of the number of users, hence, our protocols can be used even for a small number of users. In our experiments we were able to compute the differentially private median for 1 million users in 3 minutes using 3 semi-honest computation parties distributed over the Internet.

Although state-of-the-art PDF malware classifiers can be trained with almost perfect test accuracy (99%) and extremely low false positive rate (under 0.1%), it has been shown that even a simple adversary can evade them. A practically useful malware classifier must be robust against evasion attacks. However, achieving such robustness is an extremely challenging task.

In this paper, we take the first steps towards training robust PDF malware classifiers with verifiable robustness properties. For instance, a robustness property can enforce that no matter how many pages from benign documents are inserted into a PDF malware, the classifier must still classify it as malicious. We demonstrate how the worst-case behavior of a malware classifier with respect to specific robustness properties can be formally verified. Furthermore, we find that training classifiers that satisfy formally verified robustness properties can increase the evasion cost of unbounded (i.e., not bounded by the robustness properties) attackers by eliminating simple evasion attacks.

Specifically, we propose a new distance metric that operates on the PDF tree structure and specify two classes of robustness properties including subtree insertions and deletions. We utilize state-of-the-art verifiably robust training method to build robust PDF malware classifiers. Our results show that, we can achieve 92.27% average verified robust accuracy over three properties, while maintaining 99.74% accuracy and 0.56% false positive rate. With simple robustness properties, our robust model maintains 7% higher robust accuracy than all the baseline models against unrestricted whitebox attacks. Moreover, the state-of-the-art and new adaptive evolutionary attackers need up to 10 times larger $L_0$ feature distance and 21 times more PDF basic mutations (e.g., inserting and deleting objects) to evade our robust model than the baselines.

Android devices ship with pre-installed privileged apps in their firmware — some of which are essential system components, others deliver a unique user experience — that users cannot disable. These pre-installed apps are assumed to be secure as they are handpicked or developed by the device vendors themselves rather than third parties. Unfortunately, we have identified an alarming number of Android firmware that contain privilege-escalation vulnerabilities in pre-installed apps, allowing attackers to perform unauthorized actions such as executing arbitrary commands, recording the device audio and screen, and accessing personal data to name a few. To uncover these vulnerabilities, we built FIRMSCOPE, a novel static analysis system that analyzes Android firmware to expose unwanted functionality in pre-installed apps using an efficient and practical context-sensitive, flow-sensitive, field-sensitive, and partially object-sensitive taint analysis. Our experimental results demonstrate that FIRMSCOPE significantly outperforms the state-of-the-art Android taint analysis solutions both in terms of detection power and runtime performance. We used FIRMSCOPE to scan 331,342 pre-installed apps in 2,017 Android firmware images from v4.0 to v9.0 from more than 100 Android vendors. Among them, FIRMSCOPE uncovered 850 unique privilege-escalation vulnerabilities, many of which are exploitable and 0-day.

Smartphone applications that listen for network connections introduce significant security and privacy threats for users. In this paper, we focus on vetting and analyzing the security of iOS apps’ network services. To this end, we develop an efficient and scalable iOS app collection tool to download 168,951 iOS apps in the wild. We investigate a set of 1,300 apps to understand the characteristics of network service vulnerabilities, confirming 11 vulnerabilities in popular apps, such as Waze, Now, and QQBrowser. From these vulnerabilities, we create signatures for a large-scale analysis of 168,951 iOS apps, which shows that the use of certain third-party libraries listening for remote connections is a common source of vulnerable network services in 92 apps. These vulnerabilities open up the iOS device to a host of possible attacks, including data leakage, remote command execution, and denial-of-service attacks. We have disclosed identified vulnerabilities and received acknowledgments from vendors.

We propose a new class of signal injection attacks on microphones by physically converting light to sound. We show how an attacker can inject arbitrary audio signals to a target microphone by aiming an amplitude-modulated light at the microphone's aperture. We then proceed to show how this effect leads to a remote voice-command injection attack on voice-controllable systems. Examining various products that use Amazon's Alexa, Apple's Siri, Facebook's Portal, and Google Assistant, we show how to use light to obtain control over these devices at distances up to 110 meters and from two separate buildings. Next, we show that user authentication on these devices is often lacking, allowing the attacker to use light-injected voice commands to unlock the target's smartlock-protected front doors, open garage doors, shop on e-commerce websites at the target's expense, or even unlock and start various vehicles connected to the target's Google account (e.g., Tesla and Ford). Finally, we conclude with possible software and hardware defenses against our attacks.

Recently studies show that adversarial examples (AEs) can pose a serious threat to a “white-box” automatic speech recognition (ASR) system, when its machine-learning model is exposed to the adversary. Less clear is how realistic such a threat would be towards commercial devices, such as Google Home, Cortana, Echo, etc., whose models are not publicly available. Exploiting the learning model behind ASR system in black-box is challenging, due to the presence of complicated preprocessing and feature extraction even before the AEs could reach the model. Our research, however, shows that such a black-box attack is realistic. In the paper, we present Devil’s Whisper, a general adversarial attack on commercial ASR systems. Our idea is to enhance a simple local model roughly approximating the target black-box platform with a white-box model that is more advanced yet unrelated to the target. We find that these two models can effectively complement each other in predicting the target’s behavior, which enables highly transferable and generic attacks on the target. Using a novel optimization technique, we show that a local model built upon just over 1500 queries can be elevated by the open-source Kaldi Aspire Chain Model to effectively exploit commercial devices (Google Assistant, Google Home, Amazon Echo and Microsoft Cortana). For 98% of the target commands of these devices, our approach can generate at least one AE for attacking the target devices.

New advances in machine learning have made Automated Speech Recognition (ASR) systems practical and more scalable. These systems, however, pose serious privacy threats as speech is a rich source of sensitive acoustic and textual information. Although offline and open-source ASR eliminates the privacy risks, its transcription performance is inferior to that of cloud-based ASR systems, especially for real-world use cases. In this paper, we propose Prεεch, an end-to-end speech transcription system which lies at an intermediate point in the privacy-utility spectrum. It protects the acoustic features of the speakers’ voices and protects the privacy of the textual content at an improved performance relative to offline ASR. Additionally, Prεεch provides several control knobs to allow customizable utility-usability-privacy trade-off. It relies on cloud-based services to transcribe a speech file after applying a series of privacy-preserving operations on the user’s side. We perform a comprehensive evaluation of Prεεch, using diverse real-world datasets, that demonstrates its effectiveness. Prεεch provides transcription at a 2% to 32.25% (mean 17.34%) relative improvement in word error rate over Deep Speech, while fully obfuscating the speakers' voice biometrics and allowing only a differentially private view of the textual content.