In privacy-preserving cross-device federated learning, users train a global model on their local data and submit encrypted local models, while an untrusted central server aggregates the encrypted models to obtain an updated global model. Prior work has demonstrated how to verify the correctness of aggregation in such a setting. However, such verification relies on strong assumptions, such as a trusted setup among all users under unreliable network conditions, or it suffers from expensive cryptographic operations, such as bilinear pairing. In this paper, we scrutinize the verification mechanism of prior work and propose a model recovery attack, demonstrating that most local models can be leaked within a reasonable time (e.g., 98% of encrypted local models are recovered within 21 h). Then, we propose VerSA, a verifiable secure aggregation protocol for cross-device federated learning. VerSA does not require any trusted setup for verification between users while minimizing the verification cost by enabling both the central server and users to utilize only a lightweight pseudorandom generator to prove and verify the correctness of model aggregation. We experimentally confirm the efficiency of VerSA under diverse datasets, demonstrating that \textsc{VerSA} is orders of magnitude faster than verification in prior work.

Public auditing enables efficient integrity checks of data assigned to cloud servers. In this article, we revisit the public auditing for encrypted data, in which a major concern is how to effectively support data dynamics, i.e., data modification, insertion, and deletion. We first determine which factor in existing auditing schemes most limits data dynamics from a cost perspective. We then propose a novel public auditing scheme that provides data dynamics that are orders of magnitude faster than previous methods. Our auditing challenge-response protocol reduces the computation cost of the third-party auditor (TPA) significantly, thus increasing the verification speed for the auditing results. Performance and security analysis demonstrates that the proposed scheme generates minimal computation costs while guaranteeing data integrity and privacy against an untrusted cloud.

Cloud-based Internet of Things (IoT) management services are a promising means of ingesting data from globally dispersed devices. In this setting, it is important to regulate access to data managed by potentially untrusted cloud servers. Attribute-based encryption (ABE) is a highly effective tool for access control. However, applying ABE to IoT environments shows limitations in the following three aspects: First, the demands for storage resources increase in proportion to the complexity of the access control policies. Second, the computation cost of ABE is onerous for resource-limited devices. Lastly, ABE alone is intractable to prevent illegal key-sharing which leads to unauthorized access to data. In this paper, we propose an efficient and secure cloud-based IoT data management scheme using ABE. First, we remove the storage-side dependency on the complexity of the access control policies. Second, a substantial part of computationally intensive operations is securely outsourced to the cloud servers. Lastly, unauthorized access to data via illegal key-sharing is strictly forbidden. Our security analysis and experimental results show the security and practicability of the proposed scheme.

A cache side-channel attack retrieves victim’s sensitive information from a system by exploiting shared cache of CPUs. Since conventional cache side-channel attacks such as FLUSH+RELOAD and PRIME+PROBE are likely to incur numerous cache events, such as cache hits and misses, many previous strategies have focused on monitoring cache events for attack detection. However, as recently proposed attacks such as PRIME+ABORT have exploited the other events as side-channels, it has become challenging to detect them by monitoring only cache events. In this paper, we investigate PRIME+ABORT attack and identifies Intel TSX hardware events are tightly coupled with it as well as cache events. Based on our finding, we propose a novel deep learning-based cache side-channel attack detection method called FRIME . It can concurrently detect not only the conventional attacks such as FLUSH+RELOAD, PRIME+PROBE, but also PRIME+ABORT by leveraging both event types. In order to demonstrate the efficacy of our cache side-channel attack detection scheme in diverse workload conditions in the real world, we implement it using MLP, RNN, and LSTM deep learning models, demonstrating LSTM-based method outperforms the other implementations in terms of detection accuracy.

Attribute-based encryption (ABE) offers a promising solution for flexible access control over sensitive personal health records in a mobile healthcare system on top of a public cloud infrastructure. However, ABE cannot be simply applied to lightweight devices due to its substantial computation cost during decryption. This problem could be alleviated by delegating significant parts of the decryption operations to computationally powerful parties, such as cloud servers, but the correctness of the delegated computation would be at stake. Thus, previous works enabled users to validate the partial decryption by employing a cryptographic commitment or message authentication code (MAC). This paper demonstrates that the previous commitment or MAC-based schemes cannot support verifiability in the presence of potentially malevolent cloud servers. We propose two concrete attacks on previous commitment or MAC-based schemes. We propose an effective countermeasure scheme for securing resource-limited mobile healthcare systems and provide a rigorous security proof in the standard model, demonstrating that the proposed scheme is secure against our attacks. The experimental analysis shows that the proposed scheme provides the similar performance compared with the previous commitment-based schemes and outperforms the MAC-based scheme.

Many commercial cloud service providers (CSPs) adopt pay-per-query pricing models, in which data owners are charged based on the amount of data scanned by each query. In such a data sharing model, not only the privacy preservation for the data and queries but also the trustworthiness of the underlying billing system is of the utmost importance. In this paper, we revisit multi-key searchable encryption (MKSE), an efficient and secure data search algorithm allowing a data owner to grant users the ability to retrieve data of interest over the outsourced, encrypted datasets. We first investigate which factor in existing MKSE schemes renders authorized users over-privileged such that, without risking their credits (e.g., leaking the private keys and/or the passwords for their accounts associated with a project where the shared data resides), they can allow unauthorized users to make valid queries. Unfortunately, this concern may be devastating because the queries made by unauthorized users would incur unexpected financial damage to the owner in practical pay-per-query models. We then propose a novel multi-key data search scheme that is resilient to unauthorized queries. The proposed scheme features a novel user authorization mechanism that carefully limits user privilege such that even an authorized user cannot illegally invite unauthorized users to query unless he entirely leaks his credit. We demonstrate the proposed scheme is comparable to prior work in terms of performance while achieving a higher level of security.