While voiceprint authentication offers convenient user authentication and access control through voice feature recognition, a critical research gap remains: existing voiceprint authentication systems fail to simultaneously achieve sound security against replay, spoofing, and adversarial attacks, preserve voice privacy leakage, and satisfy usability demand. Previous efforts have struggled to balance these issues comprehensively. To bridge this gap, we present SeVoAuth, a cloud-based Voiceprint Authentication as a Service (VAaaS) system designed to provide privacy preservation, robust security, and enhanced usability. SeVoAuth stores a synthesized voiceprint of a user in the cloud during user registration, thereby safeguarding the privacy of the real voiceprint of the user. During user authentication, SeVoAuth applies a hash function to continuously transform features of the synthesized voiceprint, dynamically generating new verification targets for voiceprint feature mapping in each authentication session. This dynamic transformation approach effectively mitigates replay, spoofing, and adversarial attacks without requiring complex user interactions. We conduct a thorough analysis on the security and privacy of SeVoAuth and proceed to implement a prototype for performance evaluation through a series of user tests. Experimental results demonstrate that SeVoAuth outperforms cutting-edge approaches, achieving an average authentication accuracy of 99.47%, and an average Precise Detection Rate (PDR) of 98.35% against various attacks. SeVoAuth is evaluated as highly secure, efficient, and user-friendly across various circumstances.

Smart home platforms predominantly adopt the broker-mediated publish/subscribe model (e.g., MQTT) for seamless device-app coordination. However, this architecture introduces a fundamental privacy-functionality conflict: the broker requires plaintext metadata (topic strings) for message routing, which inadvertently exposes fine-grained user behavioral patterns to semi-trusted service providers. Furthermore, existing systems lack rigorous cryptographic enforcement for app permissions, leaving the ecosystem vulnerable to over-privileged or malicious apps. While conventional Attribute-Based Encryption (ABE) provides fine-grained read-side access control, it cannot enforce writer-bound policies and remains computationally prohibitive for resource-constrained IoT nodes. In this paper, we propose SHERLOC, a practical and privacy-preserving framework that reconciles secure message routing with fine-grained privilege control. SHERLOC introduces two core primitives: (1) Secret Queue Telemetry Transport (SQTT), which leverages a novel trapdoor-based matching mechanism to support multi-level wildcard routing while ensuring topic indistinguishability and resistance against inside keyword-guessing attacks (IKGA); and (2) Outsourced Inner-Product Access Control Encryption (OS-IPACE), an attribute-hiding scheme that enforces dual no-read and no-write security for apps by offloading intensive pairing operations to a local hub without compromising data secrecy. We provide formal security proofs reducing SHERLOC’s privacy guarantees to the SXDH assumption. Experimental results from a full-scale prototype, comprising ESP32-based devices and Android apps, demonstrate that SHERLOC incurs millisecond-level latency and maintains compatibility with legacy MQTT brokers without altering protocol semantics, making it a robust and deployable solution for modern smart home environments.

With the proliferation of sensitive time series data being collected, stored, and queried in cloud environments, it is imperative to integrate owner-centric privacy controls with encrypted data processing to enable secure and flexible data sharing. Existing systems, however, often lack the flexibility to express nuanced privacy policies and fail to adequately protect sensitive metadata. Adversaries can exploit declared policies, query attributes, and data access patterns to infer confidential information about data owners. In this paper, we present Dobby, a privacy-preserving time series data analytics system that enforces fine-grained and flexible access policies through function secret sharing (FSS). Dobby ensures robust privacy protection for policies, query attributes, and access patterns, while achieving malicious security in a two-party computation setting. Furthermore, we propose an optimized algorithm to streamline policy evaluation, significantly reducing communication overhead. Our evaluation demonstrates the efficiency and practicality of Dobby. For a query involving 100 ciphertexts, the system achieves a query latency of approximately 3.3 s on 10,000 data streams, each governed by an access policy comprising eight conditions.

Threshold-ECDSA is widely used to secure blockchain transactions on-chain or inter-chain thus their online performance is of great importance. The first practical thresholdECDSA was proposed by Gennaro and Goldfeder (CCS'18), based on which, Canetti et al. (CCS'20) built a threshold-ECDSA with cheater identification mechanism (CIM) and fast online signing. However, Canetti et al.'s protocol came with much more communication and computation burden and cannot apply to timely transactions. In this paper, we propose a performanceoptimized and versatile threshold-ECDSA (OptiVersa-ECDSA) with efficient CIM and fast online signing. We propose novel tools named “verifiable secret-product sharing (VSPS)” and batch VSPS, which is tailored for threshold-ECDSA. Security analysis shows that OptiVersa-ECDSA is secure against malicious adversaries in the dishonest majority model. The performance analysis shows that OptiVersa-ECDSA has 35%-65% bandwidth and runtime improvement (in particular, 99% for the CIM module), compared to Canetti et al. (CCS'20). These demonstrate that OptiVersa-ECDSA is a practical solution to secure the blockchain transactions.

Machine Learning as a Service raises the joint challenge of protecting client inputs and model parameters while enabling public verification of inference results. Existing verifiable privacy-preserving approaches often incur costly online proof generation or provide limited support for nonlinear operations. We present AuditML, a publicly auditable privacy-preserving framework for machine learning inference in the semi-honest model. AuditML combines arithmetic secret sharing with commitment-based audit data and separates execution from verification: the parties generate correlated randomness offline, perform privacy-preserving inference online, and publish the opened values and associated audit data to a public bulletin board, from which anyone can verify the computation afterward. The supported inference circuits are expressed using three auditable operations, ADD, MULTIPLY, and SIGN. In particular, AuditML implements SIGN and its underlying comparison entirely over arithmetic shares, avoiding arithmetic-to-Boolean conversion while retaining commitment-based auditability. This design eliminates a separate online proof-generation procedure and keeps auditing outside the latency-critical online phase. We implement AuditML on MP-SPDZ and evaluate linear regression, fixed-threshold binary logistic classification, and linear-kernel SVM in two-party and three-party settings. At a batch size of 50, linear and fixed-threshold binary logistic inference complete online in less than one second, while the multiplication-intensive SVM requires 10.0939–40.0334 seconds; per-participant online communication is approximately twice that of the semi-honest baseline. And compared with zkCNN-MPL, AuditML has lower inference latency with the same level of security. The results demonstrate efficient online execution for the evaluated lightweight models while exposing the higher computation and audit costs of multiplication-intensive circuits.

Encryption is the most direct technique to protect data confidentiality when users outsource their data to the cloud. Typically, ciphertexts are stored in the cloud, while cryptographic keys are managed by a key management server (KMS). However, this approach introduces new challenges in securely managing both keys and ciphertexts. Specifically, two crucial issues remain unresolved. One is that the simultaneous leakage of data encryption key and ciphertexts stored in cloud can directly compromise user data. Another is that if the cloud and KMS collude, they can trivially retrieve user data. In this work, we propose a Password-protected Encrypted Cloud Storage scheme PECS that is resilient to the aforementioned leakage and collusion attacks. In PECS, the users can encrypt/decrypt their data using only a password without storing any key material, and neither the cloud nor KMS learns any information about the user data or keys. Technically, we introduce a re-encryption mechanism performed by the cloud to prevent an adversary from obtaining the original ciphertexts. Furthermore, the user encrypts key wrap before sending it to the cloud, under a pair of password-derived secret and public key. Both the data encryption keys and password are protected from being exposed by the servers through an oblivious pseudorandom function (OPRF). Provable security and efficiency of PECS are demonstrated through comprehensive analyses.

Organic carbon flux entering the pedosphere through forest litterfall drives the spatiotemporal dynamics of soil respiration (RS). Synthesis of 14,912 in-situ observations across 843 sites parameterized a remote sensing-driven statistical model to map global forest litterfall production, PFL, at 500 m resolution (2000-2022). Global annual average PFL reached 30.06 Pg of dry mass (95% CI: 28.91-31.22 Pg). Production density exhibited an average increase of 8.25 +/- 1.37 & times; 10-3 t & sdot;ha-1 & sdot;yr-2, with upward trends spanning 50.64% (95% CI: 49.20%-52.15%) of global forest areas. Statistically significant rises occurred across 13.87% (95% CI: 12.50%-15.10%) of these domains, predominantly within tropical evergreen broadleaf and boreal needleleaf forests. Temperature functioned as the primary driver of global PFL variability, while localized environmental factors constrained regional dynamics. Causal decoupling via asymmetric residual analysis quantified the standardized sensitivity slope of RS to PFL at 0.016 (95% CI: 0.011-0.021). Implementation of Olson's first-order decay kinetics, modeling exponential substrate decomposition over time, revealed rapid tropical turnover contrasting with profound temperate biogeochemical inertia; this lag effect yielded a 24.62% explanatory gain at a one-year lag, persisting at 2.75% after four years. Global validation across 128 in-situ manipulation experiments demonstrated that asymmetric sensitivity index, defined as the ratio of respiratory log-responses to litterfall removal versus addition, shifted systematically from-0.151 in the tropics to-0.558 in temperate regions. This confirms a mechanistic transition from acute input-dependency to robust legacy-buffering along climatic gradients. Ultimately, these findings bridge fine-scale PFL-RS coupling gaps, providing critical physical constraints for global biogeochemical models.

Distributed key-value (KV) data collection under local differential privacy (LDP) faces critical challenges due to empirical parameter dependency, significant accuracy loss under high privacy intensities, and vulnerability to poisoning attacks. This paper proposes ARMKV, an automated and robust KV data collection framework featuring adaptive sampling and robust aggregation. We innovatively design a parameter adaptation mechanism that dynamically optimizes padding lengths and sampling rates based on pre-estimated data sparsity, effectively mitigating budget dilution in long-tail distributions. Furthermore, a multi-pair reporting mechanism is introduced to maximize sample utility under strict LDP constraints, coupled with an adaptive trimmed-mean aggregation strategy at the server side to filter malicious extreme values while maintaining high statistical fidelity. Theoretical analysis proves that ARMKV satisfies $\epsilon$ LDP and provides rigorous upper bounds for mean squared error. Extensive experiments on synthetic and real-world datasets (MSNBC and Jester) demonstrate that ARMKV outperforms state-of-the-art baselines like PrivKVM and PCKV, achieving an average 60% reduction in estimation error and exhibiting superior resilience against data poisoning.

Recent memory agents improve LLMs by extracting experiences and conversation history into an external storage. This enables low-overhead context assembly and online memory update without expensive LLM training. However, existing solutions remain passive and reactive; memory growth is bounded by information that happens to be available, while memory agents seldom seek external inputs in uncertainties. We propose autonomous memory agents that actively acquire, validate, and curate knowledge at a minimum cost. U-Mem materializes this idea via (i) a cost-aware knowledge-extraction cascade that escalates from cheap self/teacher signals to tool-verified research and, only when needed, expert feedback, and (ii) semantic-aware Thompson sampling to balance exploration and exploitation over memories and mitigate cold-start bias. On both verifiable and non-verifiable benchmarks, U-Mem consistently beats prior memory baselines and can surpass RL-based optimization, improving HotpotQA (Qwen2.5-7B) by 14.6 points and AIME25 (Gemini-2.5-flash) by 7.33 points.

Non-interactive group key exchange is a fundamental primitive for establishing secure communication channels in multi-party settings. However, existing protocols usually rely on computationally expensive tools, such as multi-linear maps and indistinguishability obfuscation, or lack efficient support for dynamic groups. To address these limitations, we propose a secure and efficient non-interactive group key exchange protocol built upon bilinear maps, which offers the following advantages over existing solutions. First, our protocol suits well for dynamic groups by enabling non-interactive key updates to accommodate frequent membership changes. Second, it exhibits two desirable features for distributed applications: sender un-restriction, which permits any party to send a message, and recipient un-restriction, which allows a sender to select an arbitrary subset of members as recipients. Furthermore, our protocol generates constant-size ciphertexts, enhancing its efficiency and scalability. We formally prove that our protocol achieves fully collusion-resistance in the standard model under the k-Bilinear Diffie-Hellman Exponent (k-BDHE) assumption. Performance analysis confirms the protocol's practical efficiency.

Multi-factor authentication (MFA) is extensively employed in mobile applications to enhance security, including Internet of Vehicles, healthcare systems, smart homes, etc. Traditional MFA requires users to present specific factors, which can be inconvenient if certain factors are unavailable. To address this, $ (t, n) $-threshold MFA (T-MFA) allows users to select any $ t $ out of $ n $ registered factors for authentication. However, existing T-MFA solutions face four key issues: (i) reliance on $ n-1 $ devices, which may be impractical; (ii) susceptibility to denial of service when the mandatory factor fails; (iii) limited factor types, reducing user flexibility; and (iv) increasing client-side computational costs with higher $ t $. In this work, we propose a veritable $ (t, n) $-threshold multi-factor authenticated key exchange protocol that addresses these challenges. Utilizing oblivious programmable pseudorandom functions (OPPRF) as main tools, we eliminate dependence on multiple devices, mandatory factors, and restricted factor types, achieving what we called veritable. We present a new construction of batched OPPRF to reduce client-side costs from $ O(t) $ to $ O(1) $, with 2 exponentiations cost by the client and $ t+1 $ by the server. We implement it with JavaScript to validate its flexibility and efficiency, making it highly suitable for mobile device applications.

With the proliferation of sensitive time series data being collected, stored, and queried in cloud environments, it is imperative to integrate owner-centric privacy controls with encrypted data processing to enable secure and flexible data sharing. Existing systems, however, often lack the flexibility to express nuanced privacy policies and fail to adequately protect sensitive metadata. Adversaries can exploit declared policies, query attributes, and data access patterns to infer confidential information about data owners. In this paper, we present Dobby, a privacy-preserving time series data analytics system that enforces fine-grained and flexible access policies through function secret sharing (FSS). Dobby ensures robust privacy protection for policies, query attributes, and access patterns, while achieving malicious security in a two-party computation setting. Furthermore, we propose an optimized algorithm to streamline policy evaluation, significantly reducing communication overhead. Our evaluation demonstrates the efficiency and practicality of Dobby. For a query involving 100 ciphertexts, the system achieves a query latency of approximately 3.3 s on 10,000 data streams, each governed by an access policy comprising eight conditions.

Threshold signatures are essential for fault-tolerant applications among groups of users, such as in blockchain transactions. SM2 is a digital signature standard in China and ISO, yet its threshold variant is less developed compared to international alternatives such as ECDSA. Specifically, modern threshold signatures offer identifiable abort (ID-abort) and non-interactive online signing, but these features make threshold-SM2 costly, limiting its real-world application. In this paper, we introduce a fast threshold-SM2 with ID-abort and non-interactive online signing. We design a technology for ID-abort by checking hard-to-verify pseudononces based on their mappings in a group. By putting the message-independent computations to a presigning phase, we achieve a non-interactive online signing. We prove that our threshold-SM2 is secure in the dishonest majority model and implement it using Golang. Theoretical analysis and experimental results demonstrate that our threshold-SM2 provides rich functionalities with good performance, significantly reducing computational and communication costs compared to the state-of-the-art threshold-SM2 by Liang and Chen (FCS’ 24).

Cloud-edge computing has been widely-adopted for large-scale data sharing and processing. In practical data sharing systems, data are very sensitive and typically encrypted, such as health records. Unauthorized users may attempt to decrypt ciphertexts to recover the data. Due to mistakes or malice, some users might try to share sensitive information with others who do not have access. Clearly, strong access control should be employed to restrict the read and write privilege of users. There was a rich literature on mandatory fine-grained information flow control for such scenarios, but three important issues remain. First, payload privacy was often neglected. Most of the known solutions focused on the protection ciphertext header, but ignored the payload, i.e. encrypted data, which may leak information by a malicious sender. Second, no guarantee of the encrypted data. Ill-formed ciphertexts, e.g. encrypted garbage data, can pass the global policy check, causing decryption failures or disseminating bad information, hence are incapable of content distribution. Finally, the heavy computation cost of sender authentication impedes the practical deployment. In this work, we introduce Hodor, a robust fine-grained information flow control scheme that not only guards the transmission channel with mandatory fine-grained access control for massive data, but also protects whole data traffic, checks ciphertext well-formedness, and efficiently authenticates the sender. In particular, Hodor considers full data traffic protection of both the ciphertext header and encrypted payload to resist information leakage, completely verifies the consistency between the claimed access structure and the actual access structure, and achieves efficient sender authentication with a succinct challenge-response protocol. We present a formal model and give detailed proofs. We also implement and evaluate Hodor using various optimization techniques to boost its performance. The results demonstrate the efficiency and practicality of Hodor for cloud-edge data sharing.

Blockchain wallets protect users' assets by securely storing keys. Research has explored various aspects, including BIP32 for key derivation in cold/hot settings and PDPKS further enhancing key insulation. However, most wallets are either user-managed or exchange-hosted, creating centralized points of failure. Shared-custodial wallets allow users and service providers to collaboratively manage keys, which can lead to potential corruption while still safeguarding asset security. Nonetheless, the need for server involvement in each signature poses challenges to user privacy and efficiency. To address these limitations, we propose NEST, the first efficient wallet that simultaneously satisfies strong key insulation, shared custody and privacy. We restrict malicious users' access to the server through password-based identity authentication and introduce blindness and unlinkability to prevent malicious servers from associating transactions with specific users. Additionally, we have designed two modes: interactive signing for large transactions with high security requirements, and non-interactive signing for small, high-frequency transactions. Our construction of NEST is based on bilinear pairing and an IND-CCA-secure public key encryption scheme, accompanied by a rigorous security analysis in the random oracle model. We implemented our scheme, and the results demonstrate its practicality.

Privacy-preserving machine learning algorithms are essential in federated learning, enabling model training on user data while safeguarding sensitive information. Differential privacy federated learning algorithms, such as DP-FedAvg, address this by perturbing gradients with noise proportional to their sensitivity. However, key challenges remain, notably in selecting appropriate gradient clipping thresholds and adapting fixed values to heterogeneous data distributions. In this paper, we introduce FedAdaClip, a global gradient-aware adaptive clipping algorithm for differential privacy. FedAdaClip integrates online quantile estimation and gradient perception: during early training, quantile-based methods rapidly identify suitable threshold ranges; in later stages, thresholds are dynamically adjusted based on previous gradient norms, eliminating manual tuning. Theoretical analysis demonstrates that FedAdaClip minimizes composite clipping and noise errors under (epsilon, delta)DP constraints, and guarantees convergence for non-convex objectives. Empirical results on MNIST and CIFAR-10/100 show that FedAdaClip outperforms fixed-threshold baselines such as DP-FedAvg and DP-SCAFFOLD, achieving 1.5%-7% higher test accuracy under equivalent privacy budgets. Moreover, training curves are smoother, substantially mitigating oscillations and divergence induced by suboptimal thresholds.

The breaches of the blockchain wallet keys greatly harm the security of blockchain transactions. To protect the secret keys, the known solutions, such as hierarchical deterministic wallets proposed in BIP32 or stealth addresses adopted in Monero, have been extensively researched. However, most of the existing works assume the key is safe, in the sense that it cannot be stolen or damaged, which is not true in practice. Moreover, current key revocation mechanisms either rely on centralized authorities, compromising decentralization, or require economic incentives to ensure nodes remain consistantly online. In this paper, we introduce Cocoon, the first blockchain wallet scheme that supports stealth addresses and provides a wallet revocation mechanism without the need for certificates. Cocoon not only ensures the privacy of wallet secret keys but also can individually revoke compromised keys with high performance. Our contributions are three-fold: First, we present the formal model and the related security definitions. Next, we give a generic construction based on the hierarchical identity-based signature, identity-based key encapsulation mechanism and non-interactive zero-knowledge proof. We then extend the scheme to the hierarchical setting for diverse scenarios. Finally, we give the implementation, and the results show that the scheme is practical.

Attribute-based encryption (ABE) has emerged as a new paradigm for access control in cloud computing. However, despite the many promising features of ABE, its deployment in real-world systems is still limited, partially due to the expensive cost of its underlying mathematical operations, which often grow linearly with the size and complexity of the system's security policies. This becomes particularly challenging in data-intensive applications, where multiple users may simultaneously access and manipulate large volumes of data, resulting in high levels of con- currency and demand for computing resources, which are too heavy even for high-end servers. Further exacerbating the issues are the functionality and security requirements of a cloud, as they introduce additional computations to both the client and the server. Therefore, in this work, we introduce GPABE, the first GPU-based parallelization framework for ABE to facilitate its batch processing in cloud computing. By analyzing ABE's major computational workload, we identify multiple arithmetic modules that are common in the design of pairing-based ABEs. Based on the analysis, we further propose to decompose the ABE algorithm into computation graph, which can be efficiently implemented on the GPU platform. Our graph representation bridges the gap between ABE's high-level design and their low-level implementation on GPUs, and is applicable to a variety of popular schemes in the realm of ABE. We then implement GPABE as a heterogeneous computing server, with several optimization techniques to improve its throughput. Finally, we evaluate the GPU implementation of several ABE schemes using GPABE. The results show a speedup of least 51.0x and at most 253.6x for the throughput of ABE algorithms, compared to their state-of-the-art CPU implementations, which preliminarily demonstrated the effectiveness of GPABE.

AbstractThe non-transferability of a designated confirmer signature scheme allows a signer to control the verification ability of a signature, hence protecting the signer’s privacy. However, a designated confirmer signature is insufficient when the secret keys are damaged and incapable of collaborative signature generation. In this paper, we circumvent these limitations by introducing the notion of designated confirmer threshold signature. First, we present a formal security model, then give a generic construction, which utilizes threshold signature schemes, encryption schemes and $$\Sigma$$ Σ -protocols. Instantiating this generic construction, we have two specific schemes, based on threshold Schnorr and threshold ECDSA, respectively. We further design two efficient $$\Sigma$$ Σ -protocols for efficient proofs. We also implement these schemes, and the experiment results show that our schemes are practical with rich functionalities. Finally, we demonstrate interesting applications for blockchains, such as verifiable asset auctions in blockchain and traditional electronic bidding.