Threat Of Leakage Research Articles

An SGX-based online voting protocol with maximum voter privacy

Electronic voting (E-voting), a crucial method in modern society, balances efficiency with the need for equity, reliability, and privacy to accurately record votes and maintain democracy. However, there is the challenge of potential leakage of voting information in scenarios where administrators of voting system may act maliciously. Moreover, contemporary collaborative counting methods face a significant increase in the cost of storage and heightened communication overhead in large-scale e-voting. In response, this paper proposes an innovative online voting protocol that integrates Intel SGX, homomorphic encryption, and zero-knowledge proof. This protocol aims to cut down the cost of communication and storage while maximizing the protection of voter privacy. The protocol relies on SGX to provide a trusted execution environment for the voting process, guarding against malicious attacks from external software or high-privileged administrators. In the proposed protocol, voters apply homomorphic encryption algorithms to locally encrypt their ballots, no longer relying on other entities (such as the voting system or SGX) for encryption. Then the encrypted ballots are submitted to SGX to mitigate the threat of data leakage in SGX. Through the adoption of zero-knowledge proof technology, the voting protocol can verify the legitimacy of votes without revealing their content. The proposed solution introduces the SGXVOT architecture, comprising two enclaves – Enclave V responsible for tallying individual encrypted votes, and Enclave T responsible for decrypting and publishing voting results. This design ensures that individual voters’ ballots remain in ciphertext, preserving the confidentiality of votes and voter privacy. Additionally, the protocol incorporates a “One-Time Pad” encryption communication protocol to guarantee the confidentiality of communication messages between Enclave V and Enclave T. The thorough security analysis and performance evaluation demonstrate the superior performance of the proposed solution in terms of security, practicality, and scalability.

Delegated quantum neural networks for encrypted data

Quantum machine learning is expected to utilize the potential advantages of quantum computing to advance the efficiency of machine learning. However, with the help of quantum cloud servers, ordinary users may confront the threat of privacy leakage of input data and models when performing the training or inference of quantum neural networks (QNNs). To address this problem, we present a new framework that allows the training and inference of delegated QNNs to be performed on encrypted data to protect the privacy of users’ data and models. This framework contains two models that are alternately trained: an encryptor and a predictor. The classical client first trains the encryptor defined by a classical neural network to map plaintext input data to vastly different ciphertext data. The ciphertext data is sent to the quantum cloud server to train the predictor defined by a QNN, which can indirectly predict the labels of plaintext data. With the trained encryptor and predictor, the client can send the encrypted data to the server for prediction and obtain almost equivalent prediction results. The proposed framework is applied to three types of QNN models, each dealing with low-dimensional tabular data, image data, and one-dimensional time series data, respectively. Experimental results show that the privacy protection method based on our framework can protect data and model privacy without degrading the performance of QNNs. The framework does not require users to have quantum capabilities and is suitable for protecting data and model privacy for various QNN models.

EPA-GAN: Electric Power Anonymization via Generative Adversarial Network Model

The contemporary landscape of electricity marketing data utilization is characterized by increased openness, heightened data circulation, and more intricate interaction contexts. Throughout the entire lifecycle of data, the persistent threat of leakage is ever-present. In this study, we introduce a novel electricity data anonymization model, termed EPA-GAN, which relies on table generation. In comparison to existing methodologies, our model extends the foundation of generative adversarial networks by incorporating feature encoders and feedback mechanisms. This adaptation enables the generation of anonymized data with heightened practicality and similarity to the original data, specifically tailored for mixed data types, thereby achieving a deliberate decoupling from the source data. Our proposed approach initiates by parsing the original JSON file, encoding it based on variable types and features using distinct feature encoders. Subsequently, a generative adversarial network, enhanced with information, downstream, generator losses, and the Was + GP modification, is employed to generate anonymized data. The introduction of random noise fortifies privacy protection during the data generation process. Experimental validation attests to a conspicuous reduction in both machine learning utility and statistical dissimilarity between the data synthesized by our proposed anonymization model and the original dataset. This substantiates the model’s efficacy in replacing the original data for mining analysis and data sharing, thereby effectively safeguarding the privacy of the source data.

PPML-Omics: A privacy-preserving federated machine learning method protects patients' privacy in omic data.

Modern machine learning models toward various tasks with omic data analysis give rise to threats of privacy leakage of patients involved in those datasets. Here, we proposed a secure and privacy-preserving machine learning method (PPML-Omics) by designing a decentralized differential private federated learning algorithm. We applied PPML-Omics to analyze data from three sequencing technologies and addressed the privacy concern in three major tasks of omic data under three representative deep learning models. We examined privacy breaches in depth through privacy attack experiments and demonstrated that PPML-Omics could protect patients' privacy. In each of these applications, PPML-Omics was able to outperform methods of comparison under the same level of privacy guarantee, demonstrating the versatility of the method in simultaneously balancing the privacy-preserving capability and utility in omic data analysis. Furthermore, we gave the theoretical proof of the privacy-preserving capability of PPML-Omics, suggesting the first mathematically guaranteed method with robust and generalizable empirical performance in protecting patients' privacy in omic data.

Hierarchical Perceptual Noise Injection for Social Media Fingerprint Privacy Protection.

Billions of people share images from their daily lives on social media every day. However, their biometric information (e.g., fingerprints) could be easily stolen from these images. The threat of fingerprint leakage from social media has created a strong desire to anonymize shared images while maintaining image quality, since fingerprints act as a lifelong individual biometric password. To guard the fingerprint leakage, adversarial attack that involves adding imperceptible perturbations to fingerprint images have emerged as a feasible solution. However, existing works of this kind are either weak in black-box transferability or cause the images to have an unnatural appearance. Motivated by the visual perception hierarchy (i.e., high-level perception exploits model-shared semantics that transfer well across models while low-level perception extracts primitive stimuli that result in high visual sensitivity when a suspicious stimulus is provided), we propose FingerSafe, a hierarchical perceptual protective noise injection framework to address the above mentioned problems. For black-box transferability, we inject protective noises into the fingerprint orientation field to perturb the model-shared high-level semantics (i.e., fingerprint ridges). Considering visual naturalness, we suppress the low-level local contrast stimulus by regularizing the response of the Lateral Geniculate Nucleus. Our proposed FingerSafe is the first to provide feasible fingerprint protection in both digital (up to 94.12%) and realistic scenarios (Twitter and Facebook, up to 68.75%). Our code can be found at https://github.com/nlsde-safety-team/FingerSafe.

A fully secure lattice-based signcryption with designated equality test in standard model

In hospital information systems, large volumes of electronic diagnostic records (EDRs) take up most of the storage space. While cloud server providers can reduce the local storage burden on hospitals and provide data-sharing services, the potential threat of sensitive data leakage and non-traceability of diagnoses prevents hospitals from uploading patients' diagnostic records to remote cloud servers directly. Therefore, to address security and traceability issues, we propose a new construction of post-quantum secure signcryption scheme with a designated equality test based on lattices (LB-SCDET) in this paper. Our LB-SCDET scheme allows a designated tester to perform equality tests on signcrypt-ciphertexts of EDRs without leaking what the EDRs actually are. Compared to the recent LB-SCET scheme proposed by Le et al., our LB-SCDET scheme implements a designated mechanism and is secure against offline message recovery attacks (OMRA). The comparison shows that our scheme enjoys a higher level of security, albeit the ciphertext size is slightly larger. Finally, we prove the scheme to be secure under the hardness of the Learning-with-Errors problem and unidirectionality of the Short-Integer-Solution problem. To be of independent interest, we show that the LB-SCET scheme fails to achieve the claimed IND-CCA security.

IScene: An interpretable framework with hierarchical edge services for scene risk identification in 6G internet of vehicles

AbstractScene risk identification is essential for the traffic safety of Internet of Vehicles. However, the performance of existing risk identification approaches is heavily limited by the imbalanced historical data and the poor model interpretability. Meanwhile, the large processing delay and the potential privacy leakage threat also restrict their application. In this paper, a novel risk identification model is proposed that leverages the synthetic minority over‐sampling technique nearest neighbor (SMOTEENN) method to balance between high‐risk and low‐risk data. The risk identification model has fine interpretability by using recursive feature elimination cross validation (RFECV) with the Shapley additive explanation (SHAP) to analyze the importance of different features, and further elaborately design the Focal Loss function to tackle the disparity between the difficult and easy sample learning. The proposed interpretability scene risk identification framework, named iScene, is built on the infrastructure of 6G space‐air‐ground integrated networks (SAGINs) with blockchain assistance. The model updata efficiency and privacy preservation are effectively enhanced. An elastic computing offloading algorithm is applied to minimize the system overhead under the hierarchical edge service architecture. The experimental evaluation is carried out to verify the effectiveness of the proposed risk identification framework. The results indicate that the G‐Mean value is increased by 23.4%, while the task average response delay is reduced by 21.2%, compared to that in the traditional risk identification approaches with local computing services.

A review of secure federated learning: Privacy leakage threats, protection technologies, challenges and future directions

Advances in the new generation of Internet of Things (IoT) technology are propelling the growth of intelligent industrial applications worldwide. Simultaneously, widespread adoption of artificial intelligence (AI) technologies, such as machine and deep learning, is accelerating. Traditional machine learning models rely heavily on massive amounts of data, however collecting and processing massive amounts of data generated by network-edge devices is costly and inefficient, and poses serious risks to data privacy. As a new paradigm for statistical model training in distributed edge networks, federated learning (FL) enables data to participate in federated model training without being localized. This approach can be used to solve traditional machine learning problems of low data utilization, data privacy, and information security caused by data isolation. However, the defects of the FL framework and insecure network environments cause many security and privacy leakage problems in actual application scenarios of FL. First, the concepts, classifications, and fundamental FL principles were described. Second, the mainstream privacy security issues and classification of FL were investigated. Privacy security protection techniques for FL were then identified. Finally, challenges and future research directions for the development of FL privacy security are discussed.

Review on Information Privacy Protection based on Blockchain Principle

As a distributed general ledger technology, a basic platform of smart contract, and a new distributed computing paradigm, blockchain is bound to have an impact on the development of all walks of life, and drive a new round of technology and application changes. To have an impact on the development of all walks of life and drive a new round of technology and application changes, blockchain technology improves efficiency, reduces costs, and enhances data security. However, it also faces serious information security problems. However, while block technology improves efficiency, reducing costs and improving data security, block technology also faces serious information privacy leakage problems. promote the research, development and application of blockchain in information privacy protection. This paper uses the literature analysis method to analyze the advantages and disadvantages of the blockchain in the information privacy protection from the aspects of the blockchain privacy protection This paper uses the literature analysis method to analyze the advantages and disadvantages of the blockchain in the information privacy protection from the aspects of the blockchain privacy protection , analyze the privacy leakage threat of the blockchain in the network layer, transaction layer and application layer, and summarizes the main technologies of the blockchain in the information privacy protection of data users.

Understanding and Characterizing Side Channels Exploiting Phase-Change Memories

Recent advances in non-volatile memory (NVM), together with their performance-optimized architectural schemes, position NVMs as promising building blocks for future main memory. However, the security of such techniques has not been explored. This article performs the first study on information leakage threats in phase change memories (PCM). We propose an attack framework, R-SAW, that systematically investigates side channel vulnerabilities in representative read techniques under inter-line and intra-line interleaving for multi-level cells. Our evaluation shows that the new side channels can accurately leak program secrets (e.g., crypto keys) and are extremely robust to noise. Our work highlights the need to understand microarchitecture security for emerging memory devices.

LDPORR: A localized location privacy protection method based on optimized random response

The broad use of mobile intelligent terminals with locating functions encourages the rapid development of location-based services (LBS), which are widely used in a variety of industries such as social networking, transportation, finance, and entertainment. While enjoying the convenience brought by LBS, the threat of location privacy leakage should not be overlooked. In view of the complex encoding mechanism and low availability of the current location privacy protection methods, a local differential privacy location protection method based on optimized random response is proposed in this paper. A spatial decomposition and Hilbert encoding mechanism are designed based on the Hilbert curve, which can reduce the two-dimensional location data into one-dimensional Hilbert encoding results. A local differential privacy location perturbation method based on optimized random response is proposed to improve the availability of perturbed locations and the accuracy of data aggregation. Experiments on actual location datasets prove that the proposed method can provide better location data availability and operational efficiency based on realizing local differential privacy protection of users’ location.

CMAF-IIoT: Chaotic map-based authentication framework for Industrial Internet of Things

The Industrial Internet of Things (IIoT) revolutionizes industrial production using smart devices (SMDs) deployed in IIoT environments. These SMDs collect and transmit information from target fields for analysis and can be controlled remotely. In emergency situations, real-time access to specific SMD information is crucial for prompt actions. However, ensuring secure and reliable communication between users and SMDs in untrusted channels is a significant challenge. Privacy concerns and the limitations of existing authentication frameworks further exacerbate this challenge, necessitating a lightweight cryptography-based authentication framework. This framework is essential to enable secure and reliable communication in the diverse IIoT environment while addressing privacy leakage threats and the computational constraints of IoT devices. Recently, various lightweight cryptography-based authenticated encryption (AE) schemes have been proposed to enable encryption and decryption services for resource-constrained IIoT devices. ASCON, an efficient AE scheme, offers confidentiality, integrity, and authenticity in a single encryption and decryption operation, reducing the number of cryptographic operations required for authentication framework design. This paper presents CMAF-IIoT, a chaotic map and resource-efficient AE scheme (ASCON)-based authentication framework for IIoT, addressing the aforementioned challenges. CMAF-IIoT ensures reliable communication between SMDs and users. The framework begins with user-performed local authentication on their smart devices, followed by the establishment of a session key with the SMD after mutual authentication with the gateway. Using the session key, users securely access real-time information from SMDs deployed in the IIoT environment. The security of CMAF-IIoT is validated through formal and informal security analyses. Additionally, the efficiency of CMAF-IIoT is evaluated in terms of communication, computational, and storage costs. The results indicate that CMAF-IIoT requires [6.67% to 53.33%] low storage cost, [45.13% to 65.87%] low computational cost, and [16.46% to 83.29%] low communication cost compared to contrasted authentication frameworks. These findings highlight the viability of CMAF-IIoT for the IIoT environment, as it provides resource efficiency and enhanced security features.

Securing Distributed SGD Against Gradient Leakage Threats

This paper presents a holistic approach to gradient leakage resilient distributed Stochastic Gradient Descent (SGD). <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">First</i> , we analyze two types of strategies for privacy-enhanced federated learning: (i) gradient pruning with random selection or low-rank filtering and (ii) gradient perturbation with additive random noise or differential privacy noise. We analyze the inherent limitations of these approaches and their underlying impact on privacy guarantee, model accuracy, and attack resilience. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Next</i> , we present a gradient leakage resilient approach to securing distributed SGD in federated learning, with differential privacy controlled noise as the tool. Unlike conventional methods with the per-client federated noise injection and fixed noise parameter strategy, our approach keeps track of the trend of per-example gradient updates. It makes adaptive noise injection closely aligned throughout the federated model training. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Finally</i> , we provide an empirical privacy analysis on the privacy guarantee, model utility, and attack resilience of the proposed approach. Extensive evaluation using five benchmark datasets demonstrates that our gradient leakage resilient approach can outperform the state-of-the-art methods with competitive accuracy performance, strong differential privacy guarantee, and high resilience against gradient leakage attacks. The code associated with this paper can be found: <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/git-disl/Fed-alphaCDP</uri> .

Comparison of awareness and attitudes toward artificial intelligence among Russian- and English-speaking students at Orenburg State Medical University

BACKGROUND: Artificial Intelligence (AI) is actively implemented in medicine. Since medical students are future physicians, an assessment of their awareness and attitudes toward AI is important.&#x0D; AIM: To compare the degree of awareness and attitudes toward AI among Russian-speaking students of the Orenburg State Medical University (OrSMU) from the Russian Federation and English-speaking students from the Republic of India.&#x0D; METHODS: From March 12 to 25, 2023, a voluntary anonymous survey (28 questions) was offered to OrSMU students using the Google Forms platform. An English-language version was prepared for foreign students. All responses were analyzed statistically (calculation of mean values using Likert scale, Students t-test, and Pearsons chi-square test).&#x0D; RESULTS: A total of 331 students participated in the survey, including 214 Russian-speaking and 117 English-speaking participants (127 males, 202 females, and two did not indicate gender). All participants were divided into 2 subgroups: junior (13 years, 200 participants) and senior year (46 years, 131 participants) students. The vast majority of respondents (92.3%) knew what AI is, with a higher percentage (p0.001) among Russian-speaking students (95.8%) versus English-speaking students (84.6%). Only 34.1% of Russian-speaking and 46.2% of English-speaking students (p=0.032) were aware of the possibility of using AI in medicine. A total of 28.5% and 23.4% of Russian-speaking and 44.4% and 38.5% of English-speaking respondents, respectively, were aware of the use of AI in diagnostic radiology and pathological anatomy (p=0.004). Students expressed the greatest agreement with the statement that AI will play a significant role in the development and support of medicine in the future (mean Likert scale value of 4.23). Students were the least likely to agree with the statement that AIs diagnostic abilities are superior to the human physicians clinical experience (mean of 2.84). In the case of a disagreement between the AI and the physician, 76.7% of respondents would trust the latter to make the final decision. Most respondents considered diagnostic radiology, electrocardiogram analysis, and pathological anatomy to be promising areas for the use of AI (91.3%, 71.3%, and 70.4%, respectively). For the rest of the statements about attitudes toward AI, the average values ranged from 3.63 to 4.33. Among the disadvantages of using AI, the threat of data leakage was reported. The advantages were quick data analysis and assistance in diagnosis.&#x0D; CONCLUSIONS: English-speaking students were more aware of the use of AI in medicine, whereas students from Russia showed a more positive attitude toward AI. However, in the case of a disagreement between the physician and the AI, both groups of respondents would trust the physician to make the decision.

VoiceCloak

Faced with the threat of identity leakage during voice data publishing, users are engaged in a privacy-utility dilemma when enjoying the utility of voice services. Existing machine-centric studies employ direct modification or text-based re-synthesis to de-identify users' voices but cause inconsistent audibility for human participants in emerging online communication scenarios, such as virtual meetings. In this paper, we propose a human-centric voice de-identification system, VoiceCloak, which uses adversarial examples to balance the privacy and utility of voice services. Instead of typical additive examples inducing perceivable distortions, we design a novel convolutional adversarial example that modulates perturbations into real-world room impulse responses. Benefiting from this, VoiceCloak could preserve user identity from exposure by Automatic Speaker Identification (ASI), while remaining the voice perceptual quality for non-intrusive de-identification. Moreover, VoiceCloak learns a compact speaker distribution through a conditional variational auto-encoder to synthesize diverse targets on demand. Guided by these pseudo targets, VoiceCloak constructs adversarial examples in an input-specific manner, enabling any-to-any identity transformation for robust de-identification. Experimental results show that VoiceCloak could achieve over 92% and 84% successful de-identification on mainstream ASIs and commercial systems with excellent voiceprint consistency, speech integrity, and audio quality.

Echo TEMPEST: EM Information Leakage Induced by IEMI for Electronic Devices

Electromagnetic (EM) information leakage encourages attacks, wherein the attackers passively capture and analyze EM waves that are unintentionally generated by devices. Generally, devices with weak EM emission intensities are not targeted. However, even these devices would be subject to attacks if it becomes possible to actively sense the electrical changes that occur within them when information is processed. This article demonstrates the feasibility of the information leakage threat induced by the active sensing of input impedance changes in the input/output (I/O) circuit of an integrated circuit (IC). Specifically, the changes in the input impedance when information was transmitted from the IC, were measured by irradiating the EM waves from outside the target device. This article labels the threat as Echo TEMPEST. The experiment validated Echo TEMPEST with an evaluation board that simulated the I/O circuit of the IC, UART modules, and USB keyboards. It was also demonstrated that attackers could control the distance (obtained information from the target device), depending on the intensity of the irradiated EM waves. Furthermore, we discussed countermeasure methods focusing on the conditions for executing Echo TEMPEST.

Application of Internet of Things and Blockchain in Information Security and Privacy Protection of Global Organizations

Access control data will continue to be exposed to the threat of privacy leakage even if blockchain technology currently offers a new solution for the security and privacy of the internet of things (IoT). However, its usability and privacy are not completely leveraged. This paper first discusses the IoT and blockchain technology and then examines each technology's structural models in order to address the issue of information security and privacy protection for the global organization IoT based on blockchain. Second, the information security and privacy guarantee system based on blockchain is built with ZKP and TEE at its heart after problems with zero-knowledge proof (ZKP) and trusted execution environment (TEE) in information security guarantee based on blockchain are investigated. By comparing the simulation trials, the proposed system's viability is finally confirmed. The results demonstrate that the suggested algorithm's evidence generation time is 352 ms when it reaches the experiment's highest node 28, which is clearly faster than previous techniques.

Defending Privacy Inference Attacks to Federated Learning for Intelligent IoT with Parameter Compression

Federated learning has been popularly studied with people’s increasing awareness of privacy protection. It solves the problem of privacy leakage by its ability that allows many clients to train a collaborative model without uploading local data collected by Internet of Things (IoT) devices. However, there are still threats of privacy leakage in federated learning. The privacy inference attacks can reconstruct the privacy data of other clients based on GAN from the parameters in the process of iterations for global models. In this work, we are motivated to prevent GAN-based privacy inference attacks in federated learning. Inspired by the idea of gradient compression, we propose a defense method called Federated Learning Parameter Compression (FLPC) which can reduce the sharing of information for privacy protection. It prevents attackers from recovering the private information of victims while maintaining the accuracy of the global model. Extensive experimental results demonstrated that our method is effective in the prevention of GAN-based privacy inferring attacks. In addition, based on the experimental results, we propose a norm-based metric to assess the performance of privacy-preserving.

Buoyant Flow of H2 Vs. CO2 in Storage Aquifers: Implications to Geological Screening

Summary Hydrogen will play an important role in the quest to decarbonize the world’s economy by substituting fossil fuels. In addition to the development of hydrogen generation technologies, the energy industry will need to increase hydrogen storage capacity to facilitate the development of a robust hydrogen economy. The required hydrogen storage capacity will be much larger than current hydrogen and natural gas storage capacities. There are several geological storage options for hydrogen that include depleted hydrocarbon fields and aquifers, where more research is needed until the feasibility of storing hydrogen at scale is proved. Here, we investigate the buoyant flow of H2 (as a working gas) vs. CO2 (as a cushion gas) separately in a representative storage aquifer. Buoyant flow can affect the maximum storage, capillary trapping, likelihood of leakage, and deliverability of aquifer-stored hydrogen. After building a 2D geological reservoir model initially filled with saline water, we ran numerical simulations to determine how hydrogen placed at the bottom of an aquifer might rise through the water column. The Leverett j-function is used to generate heterogeneous capillary entry pressure fields that correlate with porosity and permeability fields. Hydrogen viscosities were based on the Jossi et al. correlation, and the density was modeled using the Peng-Robinson equation of state. We then simulated several scenarios to assess flow during short- (annually) and long- (several years) term storage. For comparison purposes, we also ran CO2 storage simulations using the same geological model but with CO2-brine-rock properties collected from the literature. For a representative storage aquifer (323 K, 15.7 MPa, and mean permeability of 200 md), significant fingering occurred as the hydrogen rose through the saline water column. The hydrogen experienced more buoyant flow and created flow paths with increased fingering when compared with CO2. Individual hydrogen fingers are thinner than the CO2 fingers in the simulations, and the tips of hydrogen finger fronts propagated upward roughly twice as fast as the CO2 front for a typical set of heterogeneity indicators (Dykstra-Parson’s coefficient Vdp = 0.80, and dimensionless autocorrelation length λDx = 2). The implications of buoyant flow for hydrogen in saline aquifers include an increased threat of leakage, more residual trapping of hydrogen, and, therefore, the need to focus more on the heterogeneity and lateral correlation behavior of the repository. If hydrogen penetrates the caprock of an aquifer, it will leak faster than CO2 and generate more vertical flow pathways. We identify possible depositional environments for clastic aquifers that would offer suitable characteristics for storage.

Off-Platform Threats in On-Demand Services

Problem definition: Online platforms that provide on-demand services are often threatened by the phenomenon of leakage, where customer-provider pairs may decide to transact “off-platform” to avoid paying commissions to the platform. This paper investigates properties of services that make them vulnerable or resistant to leakage. Academic/practical relevance: In practice, much attention has been given to platform leakage, with platforms experimenting with multiple approaches to alleviate leakage and maintain their customer and provider bases. Yet, there is a current dearth of studies in the operations literature that systematically analyze the key factors behind platform leakage. Our work fills this gap and answers practical questions regarding the sustainability of platform. Methodology: We develop two game-theoretical models that capture service providers’ and customers’ decisions whether to conduct transactions on or off the platform. In the first (“perfect information”) model, we assume that customers are equipped with information to select their desired providers on the platform, whereas in the second (“imperfect information”) model, we assume customers are randomly matched with available providers by the platform. Results: For profit maximizing platforms, we show that leakage occurs if and only if the value of the counterparty risk from off-platform transactions exceeds a threshold. Across both models, platforms tend to be more immunized against leakage as provider pool sizes increase, customer valuations for service increase, their waiting costs decrease, or variability in service times are reduced. Finally, by comparing the degree of leakage between both settings, we find that neither model dominates the other across all parameter combinations. Managerial implications: Our results provide guidance to existing platform managers or entrepreneurs who are considering “platforming” their services. Namely, based on a few key features of the operating environment, managers can assess the severity of the threat of platform leakage for their specific business context. Our results also suggest how redesigning the waiting process, reducing service time variability, upskilling providers can reduce the threat of leakage. They also suggest the conditions under which revealing provider quality information to customers can help to curb leakage. Funding: J. Goh’s work was supported by a National University of Singapore Start-Up [Grant R-314-000-110-133] and a 2021 Humanities and Social Sciences Fellowship from the National University of Singapore. Supplemental Material: The online appendices are available at https://doi.org/10.1287/msom.2022.1179 .