TLDR: Large Language Models have revolutionized AI but face growing adversarial threats. We present the adversarial landscape from a novel perspective by categorizing these threats into four key objectives: privacy breaches, integrity compromises, adversarial misuse, and availability disruptions.

TLDR: BulletCT is a new Ring Confidential Transaction (RingCT) signature scheme in the discrete logarithm setting that does not require a trusted setup. It achieves greater scalability than state-of-the-art RingCT schemes. BulletCT features a novel K-out-of-N proof for strong anonymity and a tag proof that leverages permutation constraints to achieve linkability. Additionally, we identify key limitations in applying Any-out-of-N proofs to RingCT and address a critical flaw in prior constructions.

TLDR: FlashSwift is a logarithmic-sized zero-knowledge range argument in the discrete logarithm setting without using a trusted setup. By combining the techniques from both Flashproof and SwiftRange, FlashSwift inherits and capitalizes on their efficiency advantages. It creates two new records of the smallest proof sizes, 289 bytes and 417 bytes, for 8-bit and 16-bit ranges among all the bit-decomposition-based range proofs without requiring trusted setups. Moreover, it is the first configurable range proof that is adaptable to various scenarios with different specifications.

TLDR: SwiftRange is a logarithmic-sized zero-knowledge range argument in the discrete logarithm setting without using a trusted setup. It is tailored for confidential transactions (CT) on blockchain platforms. It proves a committed value lies in the range [0, 2^N-1], where N is the bit length of the range size. It achieves double verifier efficiency of Bulletproofs with only a slightly higher communication cost for CT-friendly ranges, where N is 32 or 64.

TLDR: Flashproofs are efficient zero-knowledge proofs of knowledge in the discrete logarithm setting without using a trusted setup, consisting of:

• • • A range argument proves a committed value lies in the range [0, 2^N-1], where N is the bit length of the range size. It achieves O(N^(2/3)) efficiency in both communication and verification. Especially, the high verification efficiency makes it a suitable candidate for smart-contract blockchain platforms, whose verification consumes comparable gas costs to that of the most efficient zk-SNARK (Groth16) that relies on a trusted setup.

    • A polynomial evaluation argument proves that two committed values satisfy a public polynomial relation. It achieves logarithmic efficiency in both communication and verification and is a crucial building block for zero-knowledge arguments of membership and non-membership, where an argument of membership or non-membership proves a committed value belongs or does not belong to a public set of values. 

TLDR: We propose a multi-party computation protocol based on Paillier threshold cryptosystem. Our protocol enables privacy-preserving collaborative consumption in a semi-honest setting. 

TLDR: We propose an efficient multi-party computation protocol and a blockchain-based cost-sharing solution to achieve energy storage sharing in a privacy-preserving manner. Our protocol leverages the popular SPDZ secret-sharing framework to defend against a dishonest majority, who may arbitrarily deviate from the protocol.

TLDR: We present the first zero-knowledge framework for verifiable machine unlearning, specifically designed to support multi-granularity and forgery-resistance. Our solution enables not only traditional sample-level unlearning but also more advanced feature-level and class-level unlearning. Furthermore, we propose the first effective strategies to resist state-of-the-art forging attacks, where in stochastic gradient descent optimization, gradients from unlearned data, or from minibatches containing it, can be forged using alternative data samples or minibatches that exclude it.

TLDR: We propose the first privacy-aware, multi-dimensional framework for Email Phishing CAmpaign DEtection to automatically identify email phishing campaigns by clustering seemingly unrelated attacks. Our framework employs a hierarchical architecture combining a structural layer and a contextual layer, offering a comprehensive analysis of phishing attacks by thoroughly examining both structural and contextual elements.

TLDR: We propose an efficient multi-party computation protocol to compute ranked-based statistics, e.g., median, percentiles, over distributed datasets. Our protocol achieves higher accuracy and stronger security compared with the state-of-the-art. Moreover, we leverage different zero-knowledge proofs to defend against malicious parties from dishonestly deviating from the protocol.