With tools like GitHub Copilot, automatic code suggestion is no longer a dream in software engineering. These tools, based on large language models, are typically trained on massive corpora of code mined from unvetted public sources. As a result, these models are susceptible to data poisoning attacks where an adversary manipulates the model’s training by injecting malicious data. Poisoning attacks could be designed to influence the model’s suggestions at run time for chosen contexts, such as inducing the model into suggesting insecure code payloads. To achieve this, prior attacks explicitly inject the insecure code payload into the training data, making the poison data detectable by static analysis tools that can remove such malicious data from the training set. In this work, we demonstrate two novel attacks, Covert and TrojanPuzzle, that can bypass static analysis by planting malicious poison data in out-of-context regions such as docstrings. Our most novel attack, TrojanPuzzle, goes one step further in generating less suspicious poison data by never explicitly including certain (suspicious) parts of the payload in the poison data, while still inducing a model that suggests the entire payload when completing code (i.e., outside docstrings). This makes TrojanPuzzle robust against signature-based dataset-cleansing methods that can filter out suspicious sequences from the training data. Our evaluation against models of two sizes demonstrates that both Covert and TrojanPuzzle have significant implications for practitioners when selecting code used to train or tune code-suggestion models.

Pre-trained large language models, such as GPT-2 and BERT, are often fine-tuned to achieve state-of-the-art performance on a downstream task. One natural example is the "Smart Reply" application where a pre-trained model is tuned to provide suggested responses for a given query message. Since the tuning data is often sensitive data such as emails or chat transcripts, it is important to understand and mitigate the risk that the model leaks its tuning data. We investigate potential information leakage vulnerabilities in a typical Smart Reply pipeline. We consider a realistic setting where the adversary can only interact with the underlying model through a frontend interface that constrains what types of queries can be sent to the model. Previous attacks do not work in these settings, but require the ability to send unconstrained queries directly to the model. Even when there are no constraints on the queries, previous attacks typically require thousands, or even millions, of queries to extract useful information, while our attacks can extract sensitive data in just a handful of queries. We introduce a new type of active extraction attack that exploits canonical patterns in text containing sensitive data. We show experimentally that it is possible for an adversary to extract sensitive user information present in the training data, even in realistic settings where all interactions with the model must go through a front-end that limits the types of queries. We explore potential mitigation strategies and demonstrate empirically how differential privacy appears to be a reasonably effective defense mechanism to such pattern extraction attacks.

Designing privacy-respecting and human-centric smart cities requires a careful investigation of people’s attitudes and concerns toward city-wide data collection scenarios. To capture a holistic view, we carried out this investigation in two phases. We first surfaced people’s understanding, concerns, and expectations toward smart city scenarios by conducting 21 semi-structured interviews with people in underserved communities. We complemented this in-depth qualitative study with a 348-participant online survey of the general population to quantify the significance of smart city factors (e.g., type of collected data) on attitudes and concerns. Depending on demographics, privacy and ethics were the two most common types of concerns among participants. We found the type of collected data to have the most and the retention time to have the least impact on participants’ perceptions and concerns about smart cities. We highlight key takeaways and recommendations for city stakeholders to consider when designing inclusive and protective smart cities.

It is known that fully homomorphic encryption (FHE) can be used to build efficient (labeled) Private Set Intersection protocols in the unbalanced setting, where one of the sets is much larger than the other~(Chen et al. (CCS'17, CCS'18)). In this paper we demonstrate multiple algorithmic improvements upon these works. In particular, our protocol has an asymptotically better computation cost, requiring only O(√|X| ) homomorphic multiplications, and communication complexity sublinear in the larger set size|X|. We demonstrate that our protocol is significantly better than that of Chen et al. (CCS'18) for many practical parameters, especially in terms of online communication cost. For example, when intersecting $228 and 2048 item sets, our protocol reduces the online computation time by more than 71% and communication by more than 63%. When intersecting 224 and 4096 item sets, our protocol reduces the online computation time by 27% and communication by 63%. Our comparison to other state-of-the-art unbalanced PSI protocols shows that our protocol has the best total communication complexity when |X| ≥ 224. For labeled PSI our protocol also outperforms Chen et al. (CCS'18). When intersecting 220 and 256 item sets, with the larger set having associated 288-byte labels, our protocol reduces the online computation time by more than 67% and communication by 34%. Finally, we demonstrate a modification that results in nearly constant communication cost in the larger set size |X|, but impractically high computation complexity on today's CPUs. For example, to intersect a 210-item set with sets of size 222, 224, or 226, our proof-of-concept implementation requires only 0.76 MB of online communication, which is more than a 24-fold improvement over Chen et al. (CCS'18).

With the rapid increase in cloud computing, concerns surrounding data privacy, security, and confidentiality also have been increased significantly. Not only cloud providers are susceptible to internal and external hacks, but also in some scenarios, data owners cannot outsource the computation due to privacy laws such as GDPR, HIPAA, or CCPA. Fully Homomorphic Encryption (FHE) is a groundbreaking invention in cryptography that, unlike traditional cryptosystems, enables computation on encrypted data without ever decrypting it. However, the most critical obstacle in deploying FHE at large-scale is the enormous computation overhead. In this paper, we present HEAX, a novel hardware architecture for FHE that achieves unprecedented performance improvements. HEAX leverages multiple levels of parallelism, ranging from ciphertext-level to fine-grained modular arithmetic level. Our first contribution is a new highly-parallelizable architecture for number-theoretic transform (NTT) which can be of independent interest as NTT is frequently used in many lattice-based cryptography systems. Building on top of NTT engine, we design a novel architecture for computation on homomorphically encrypted data. Our implementation on reconfigurable hardware demonstrates 164-268× performance improvement for a wide range of FHE parameters.

Fully-Homomorphic Encryption (FHE) offers powerful capabilities by enabling secure offloading of both storage and computation, and recent innovations in schemes and implementations have made it all the more attractive. At the same time, FHE is notoriously hard to use with a very constrained programming model, a very unusual performance profile, and many cryptographic constraints. Existing compilers for FHE either target simpler but less efficient FHE schemes or only support specific domains where they can rely on expert-provided high-level runtimes to hide complications.

Homomorphic Encryption (HE) is a cryptosystem which supports computation on encrypted data. Ló pez-Alt et al. (STOC 2012) proposed a generalized notion of HE, called Multi-Key Homomorphic Encryption (MKHE), which is capable of performing arithmetic operations on ciphertexts encrypted under different keys. In this paper, we present multi-key variants of two HE schemes with packed ciphertexts. We present new relinearization algorithms which are simpler and faster than previous method by Chen et al. (TCC 2017). We then generalize the bootstrapping techniques for HE to obtain multi-key fully homomorphic encryption schemes. We provide a proof-of-concept implementation of both MKHE schemes using Microsoft SEAL. For example, when the dimension of base ring is 8192, homomorphic multiplication between multi-key BFV (resp. CKKS) ciphertexts associated with four parties followed by a relinearization takes about 116 (resp. 67) milliseconds. Our MKHE schemes have a wide range of applications in secure computation between multiple data providers. As a benchmark, we homomorphically classify an image using a pre-trained neural network model, where input data and model are encrypted under different keys. Our implementation takes about 1.8 seconds to evaluate one convolutional layer followed by two fully connected layers on an encrypted image from the MNIST dataset.

In 2009, Craig Gentry introduced the first "fully" homomorphic encryption scheme allowing arbitrary circuits to be evaluated on encrypted data. Homomorphic encryption is a very powerful cryptographic primitive, though it has often been viewed by practitioners as too inefficient for practical applications. However, the performance of these encryption schemes has come a long way from that of Gentry's original work: there are now several well-maintained libraries implementing homomorphic encryption schemes and protocols demonstrating impressive performance results, alongside an ongoing standardization effort by the community. In this tutorial we survey the existing homomorphic encryption landscape, providing both a general overview of the state of the art, as well as a deeper dive into several of the existing libraries. We aim to provide a thorough introduction to homomorphic encryption accessible by the broader computer security community. Several of the presenters are core developers of well-known publicly available homomorphic encryption libraries, and organizers of the homomorphic encryption standardization effort \hrefhttp://homomorphicencryption.org/. This tutorial is targeted at application developers, security researchers, privacy engineers, graduate students, and anyone else interested in learning the basics of modern homomorphic encryption.The tutorial is divided into two parts: Part I is accessible by everyone comfortable with basic college-level math; Part II will cover more advanced topics, including descriptions of some of the different homomorphic encryption schemes and libraries, concrete example applications and code samples, and a deeper discussion on implementation challenges. Part II requires the audience to be familiar with modern C++.

$\sf{ NTRUEncrypt}$is one of the most promising candidates for quantum-safe cryptography. In this paper, we focus on the$\sf{ NTRU743}$parameter set. We give a report on all known attacks against this parameter set and show that it delivers 256 bits of security against classical attackers and 128 bits of security against quantum attackers. We then present a parameter-dependent optimization using a tailored hierarchy of multiplication algorithms as well as the Intel AVX2 instructions, and show that this optimization is constant-time. Our implementation is two to three times faster than the reference implementation of$\sf{ NTRUEncrypt}$.

In this paper, we report on our implementation of a lattice-based key-policy attribute-based encryption (KP-ABE) scheme, which uses short secret keys. The particular KP-ABE scheme can be used directly for attribute-based access control applications, as well as a building block in more involved applications and cryptographic schemes, such as audit log encryption, targeted broadcast encryption, functional encryption, and program obfuscation. We adapt a recently proposed KP-ABE scheme based on the learning with errors (LWE) problem to a more efficient scheme based on the ring learning with errors (RLWE) problem, and demonstrate an implementation that can be used in practical applications. Our state-of-the-art implementation on graphics processing units shows that the homomorphic public key and ciphertext evaluation operations, which dominate the execution time of the KP-ABE scheme, can be performed in a reasonably short amount of time. Our practicality results also hold when scaled to a relatively large number of attributes. To the best of our knowledge, this is the first KP-ABE implementation that supports both ciphertext and public key homomorphism, and the only experimental practicality results reported in this paper.