CyberSec Research

Traditional backdoor attacks in federated learning (FL) operate within constrained attack scenarios, as they depend on visible triggers and require physical modifications to the target object, which limits their practicality. To address this limitation, we introduce a novel backdoor attack prototype for FL called the out-of-distribution (OOD) backdoor attack ($\mathtt{OBA}$), which uses OOD data as both poisoned samples and triggers simultaneously. Our approach significantly broadens the scope of backdoor attack scenarios in FL. To improve the stealthiness of $\mathtt{OBA}$, we propose $\mathtt{SoDa}$, which regularizes both the magnitude and direction of malicious local models during local training, aligning them closely with their benign versions to evade detection. Empirical results demonstrate that $\mathtt{OBA}$ effectively circumvents state-of-the-art defenses while maintaining high accuracy on the main task. To address this security vulnerability in the FL system, we introduce $\mathtt{BNGuard}$, a new server-side defense method tailored against $\mathtt{SoDa}$. $\mathtt{BNGuard}$ leverages the observation that OOD data causes significant deviations in the running statistics of batch normalization layers. This allows $\mathtt{BNGuard}$ to identify malicious model updates and exclude them from aggregation, thereby enhancing the backdoor robustness of FL. Extensive experiments across various settings show the effectiveness of $\mathtt{BNGuard}$ on defending against $\mathtt{SoDa}$. The code is available at https://github.com/JiiahaoXU/SoDa-BNGuard.

Software Bills of Materials (SBOMs) have become a regulatory requirement for improving software supply chain security and trust by means of transparency regarding components that make up software artifacts. However, enterprise and regulated software vendors commonly wish to restrict who can view confidential software metadata recorded in their SBOMs due to intellectual property or security vulnerability information. To address this tension between transparency and confidentiality, we propose Petra, an SBOM exchange system that empowers software vendors to interoperably compose and distribute redacted SBOM data using selective encryption. Petra enables software consumers to search redacted SBOMs for answers to specific security questions without revealing information they are not authorized to access. Petra leverages a format-agnostic, tamper-evident SBOM representation to generate efficient and confidentiality-preserving integrity proofs, allowing interested parties to cryptographically audit and establish trust in redacted SBOMs. Exchanging redacted SBOMs in our Petra prototype requires less than 1 extra KB per SBOM, and SBOM decryption account for at most 1% of the performance overhead during an SBOM query.

Motivated by software maintenance and the more recent concept of security debt, the paper presents a time series analysis of vulnerability patching of Red Hat's products and components between 1999 and 2024. According to the results based on segmented regression analysis, the amounts of vulnerable products and components have not been stable; a linear trend describes many of the series well. Nor do the amounts align well with trends characterizing vulnerabilities in general. There are also visible breakpoints indicating that the linear trend is not universally applicable and that the growing security debt may be stabilizing.

Synthetic data generation plays an important role in enabling data sharing, particularly in sensitive domains like healthcare and finance. Recent advances in diffusion models have made it possible to generate realistic, high-quality tabular data, but they may also memorize training records and leak sensitive information. Membership inference attacks (MIAs) exploit this vulnerability by determining whether a record was used in training. While MIAs have been studied in images and text, their use against tabular diffusion models remains underexplored despite the unique risks of structured attributes and limited record diversity. In this paper, we introduce MIAEPT, Membership Inference Attack via Error Prediction for Tabular Data, a novel black-box attack specifically designed to target tabular diffusion models. MIA-EPT constructs errorbased feature vectors by masking and reconstructing attributes of target records, disclosing membership signals based on how well these attributes are predicted. MIA-EPT operates without access to the internal components of the generative model, relying only on its synthetic data output, and was shown to generalize across multiple state-of-the-art diffusion models. We validate MIA-EPT on three diffusion-based synthesizers, achieving AUC-ROC scores of up to 0.599 and TPR@10% FPR values of 22.0% in our internal tests. Under the MIDST 2025 competition conditions, MIA-EPT achieved second place in the Black-box Multi-Table track (TPR@10% FPR = 20.0%). These results demonstrate that our method can uncover substantial membership leakage in synthetic tabular data, challenging the assumption that synthetic data is inherently privacy-preserving. Our code is publicly available at https://github.com/eyalgerman/MIA-EPT.

This work introduces xOffense, an AI-driven, multi-agent penetration testing framework that shifts the process from labor-intensive, expert-driven manual efforts to fully automated, machine-executable workflows capable of scaling seamlessly with computational infrastructure. At its core, xOffense leverages a fine-tuned, mid-scale open-source LLM (Qwen3-32B) to drive reasoning and decision-making in penetration testing. The framework assigns specialized agents to reconnaissance, vulnerability scanning, and exploitation, with an orchestration layer ensuring seamless coordination across phases. Fine-tuning on Chain-of-Thought penetration testing data further enables the model to generate precise tool commands and perform consistent multi-step reasoning. We evaluate xOffense on two rigorous benchmarks: AutoPenBench and AI-Pentest-Benchmark. The results demonstrate that xOffense consistently outperforms contemporary methods, achieving a sub-task completion rate of 79.17%, decisively surpassing leading systems such as VulnBot and PentestGPT. These findings highlight the potential of domain-adapted mid-scale LLMs, when embedded within structured multi-agent orchestration, to deliver superior, cost-efficient, and reproducible solutions for autonomous penetration testing.

Security-critical machine-learning (ML) systems, such as face-recognition systems, are susceptible to adversarial examples, including real-world physically realizable attacks. Various means to boost ML's adversarial robustness have been proposed; however, they typically induce unfair robustness: It is often easier to attack from certain classes or groups than from others. Several techniques have been developed to improve adversarial robustness while seeking perfect fairness between classes. Yet, prior work has focused on settings where security and fairness are less critical. Our insight is that achieving perfect parity in realistic fairness-critical tasks, such as face recognition, is often infeasible -- some classes may be highly similar, leading to more misclassifications between them. Instead, we suggest that seeking symmetry -- i.e., attacks from class $i$ to $j$ would be as successful as from $j$ to $i$ -- is more tractable. Intuitively, symmetry is a desirable because class resemblance is a symmetric relation in most domains. Additionally, as we prove theoretically, symmetry between individuals induces symmetry between any set of sub-groups, in contrast to other fairness notions where group-fairness is often elusive. We develop Sy-FAR, a technique to encourage symmetry while also optimizing adversarial robustness and extensively evaluate it using five datasets, with three model architectures, including against targeted and untargeted realistic attacks. The results show Sy-FAR significantly improves fair adversarial robustness compared to state-of-the-art methods. Moreover, we find that Sy-FAR is faster and more consistent across runs. Notably, Sy-FAR also ameliorates another type of unfairness we discover in this work -- target classes that adversarial examples are likely to be classified into become significantly less vulnerable after inducing symmetry.

Large Language Models (LLMs) are increasingly deployed for task automation and content generation, yet their safety mechanisms remain vulnerable to circumvention through different jailbreaking techniques. In this paper, we introduce \textit{Content Concretization} (CC), a novel jailbreaking technique that iteratively transforms abstract malicious requests into concrete, executable implementations. CC is a two-stage process: first, generating initial LLM responses using lower-tier, less constrained safety filters models, then refining them through higher-tier models that process both the preliminary output and original prompt. We evaluate our technique using 350 cybersecurity-specific prompts, demonstrating substantial improvements in jailbreak Success Rates (SRs), increasing from 7\% (no refinements) to 62\% after three refinement iterations, while maintaining a cost of 7.5\textcent~per prompt. Comparative A/B testing across nine different LLM evaluators confirms that outputs from additional refinement steps are consistently rated as more malicious and technically superior. Moreover, manual code analysis reveals that generated outputs execute with minimal modification, although optimal deployment typically requires target-specific fine-tuning. With eventual improved harmful code generation, these results highlight critical vulnerabilities in current LLM safety frameworks.

SNOVA is a post-quantum cryptographic signature scheme known for its efficiency and compact key sizes, making it a second-round candidate in the NIST post-quantum cryptography standardization process. This paper presents a comprehensive fault analysis of SNOVA, focusing on both permanent and transient faults during signature generation. We introduce several fault injection strategies that exploit SNOVA's structure to recover partial or complete secret keys with limited faulty signatures. Our analysis reveals that as few as 22 to 68 faulty signatures, depending on the security level, can suffice for key recovery. We propose a novel fault-assisted reconciliation attack, demonstrating its effectiveness in extracting the secret key space via solving a quadratic polynomial system. Simulations show transient faults in key signature generation steps can significantly compromise SNOVA's security. To address these vulnerabilities, we propose a lightweight countermeasure to reduce the success of fault attacks without adding significant overhead. Our results highlight the importance of fault-resistant mechanisms in post-quantum cryptographic schemes like SNOVA to ensure robustness.

Code-generating Large Language Models (LLMs) significantly accelerate software development. However, their frequent generation of insecure code presents serious risks. We present a comprehensive evaluation of seven parameter-efficient fine-tuning (PEFT) techniques, demonstrating substantial gains in secure code generation without compromising functionality. Our research identifies prompt-tuning as the most effective PEFT method, achieving an 80.86% Overall-Secure-Rate on CodeGen2 16B, a 13.5-point improvement over the 67.28% baseline. Optimizing decoding strategies through sampling temperature further elevated security to 87.65%. This equates to a reduction of approximately 203,700 vulnerable code snippets per million generated. Moreover, prompt and prefix tuning increase robustness against poisoning attacks in our TrojanPuzzle evaluation, with strong performance against CWE-79 and CWE-502 attack vectors. Our findings generalize across Python and Java, confirming prompt-tuning's consistent effectiveness. This study provides essential insights and practical guidance for building more resilient software systems with LLMs.

Federated Learning (FL) enables collaborative model training across decentralised clients while keeping local data private, making it a widely adopted privacy-enhancing technology (PET). Despite its privacy benefits, FL remains vulnerable to privacy attacks, including those targeting specific clients. In this paper, we study an underexplored threat where a dishonest orchestrator intentionally manipulates the aggregation process to induce targeted overfitting in the local models of specific clients. Whereas many studies in this area predominantly focus on reducing the amount of information leakage during training, we focus on enabling an early client-side detection of targeted overfitting, thereby allowing clients to disengage before significant harm occurs. In line with this, we propose three detection techniques - (a) label flipping, (b) backdoor trigger injection, and (c) model fingerprinting - that enable clients to verify the integrity of the global aggregation. We evaluated our methods on multiple datasets under different attack scenarios. Our results show that the three methods reliably detect targeted overfitting induced by the orchestrator, but they differ in terms of computational complexity, detection latency, and false-positive rates.

Simulating hostile attacks of physical autonomous systems can be a useful tool to examine their robustness to attack and inform vulnerability-aware design. In this work, we examine this through the lens of multi-robot patrol, by presenting a machine learning-based adversary model that observes robot patrol behavior in order to attempt to gain undetected access to a secure environment within a limited time duration. Such a model allows for evaluation of a patrol system against a realistic potential adversary, offering insight into future patrol strategy design. We show that our new model outperforms existing baselines, thus providing a more stringent test, and examine its performance against multiple leading decentralized multi-robot patrol strategies.

Federated Learning (FL) allows collaborative model training across distributed clients without sharing raw data, thus preserving privacy. However, the system remains vulnerable to privacy leakage from gradient updates and Byzantine attacks from malicious clients. Existing solutions face a critical trade-off among privacy preservation, Byzantine robustness, and computational efficiency. We propose a novel scheme that effectively balances these competing objectives by integrating homomorphic encryption with dimension compression based on the Johnson-Lindenstrauss transformation. Our approach employs a dual-server architecture that enables secure Byzantine defense in the ciphertext domain while dramatically reducing computational overhead through gradient compression. The dimension compression technique preserves the geometric relationships necessary for Byzantine defence while reducing computation complexity from $O(dn)$ to $O(kn)$ cryptographic operations, where $k \ll d$. Extensive experiments across diverse datasets demonstrate that our approach maintains model accuracy comparable to non-private FL while effectively defending against Byzantine clients comprising up to $40\%$ of the network.

Safety alignment is critical for the ethical deployment of large language models (LLMs), guiding them to avoid generating harmful or unethical content. Current alignment techniques, such as supervised fine-tuning and reinforcement learning from human feedback, remain fragile and can be bypassed by carefully crafted adversarial prompts. Unfortunately, such attacks rely on trial and error, lack generalizability across models, and are constrained by scalability and reliability. This paper presents NeuroStrike, a novel and generalizable attack framework that exploits a fundamental vulnerability introduced by alignment techniques: the reliance on sparse, specialized safety neurons responsible for detecting and suppressing harmful inputs. We apply NeuroStrike to both white-box and black-box settings: In the white-box setting, NeuroStrike identifies safety neurons through feedforward activation analysis and prunes them during inference to disable safety mechanisms. In the black-box setting, we propose the first LLM profiling attack, which leverages safety neuron transferability by training adversarial prompt generators on open-weight surrogate models and then deploying them against black-box and proprietary targets. We evaluate NeuroStrike on over 20 open-weight LLMs from major LLM developers. By removing less than 0.6% of neurons in targeted layers, NeuroStrike achieves an average attack success rate (ASR) of 76.9% using only vanilla malicious prompts. Moreover, Neurostrike generalizes to four multimodal LLMs with 100% ASR on unsafe image inputs. Safety neurons transfer effectively across architectures, raising ASR to 78.5% on 11 fine-tuned models and 77.7% on five distilled models. The black-box LLM profiling attack achieves an average ASR of 63.7% across five black-box models, including the Google Gemini family.

Path MTU Discovery (PMTUD) and IP address sharing are integral aspects of modern Internet infrastructure. In this paper, we investigate the security vulnerabilities associated with PMTUD within the context of prevalent IP address sharing practices. We reveal that PMTUD is inadequately designed to handle IP address sharing, creating vulnerabilities that attackers can exploit to perform off-path TCP hijacking attacks. We demonstrate that by observing the path MTU value determined by a server for a public IP address (shared among multiple devices), an off-path attacker on the Internet, in collaboration with a malicious device, can infer the sequence numbers of TCP connections established by other legitimate devices sharing the same IP address. This vulnerability enables the attacker to perform off-path TCP hijacking attacks, significantly compromising the security of the affected TCP connections. Our attack involves first identifying a target TCP connection originating from the shared IP address, followed by inferring the sequence numbers of the identified connection. We thoroughly assess the impacts of our attack under various network configurations. Experimental results reveal that the attack can be executed within an average time of 220 seconds, achieving a success rate of 70%.Case studies, including SSH DoS, FTP traffic poisoning, and HTTP injection, highlight the threat it poses to various applications. Additionally, we evaluate our attack across 50 real-world networks with IP address sharing--including public Wi-Fi, VPNs, and 5G--and find 38 vulnerable. Finally, we responsibly disclose the vulnerabilities, receive recognition from organizations such as IETF, Linux, and Cisco, and propose our countermeasures.

A red team simulates adversary attacks to help defenders find effective strategies to defend their systems in a real-world operational setting. As more enterprise systems adopt AI, red-teaming will need to evolve to address the unique vulnerabilities and risks posed by AI systems. We take the position that AI systems can be more effectively red-teamed if AI red-teaming is recognized as a domain-specific evolution of cyber red-teaming. Specifically, we argue that existing Cyber Red Teams who adopt this framing will be able to better evaluate systems with AI components by recognizing that AI poses new risks, has new failure modes to exploit, and often contains unpatchable bugs that re-prioritize disclosure and mitigation strategies. Similarly, adopting a cybersecurity framing will allow existing AI Red Teams to leverage a well-tested structure to emulate realistic adversaries, promote mutual accountability with formal rules of engagement, and provide a pattern to mature the tooling necessary for repeatable, scalable engagements. In these ways, the merging of AI and Cyber Red Teams will create a robust security ecosystem and best position the community to adapt to the rapidly changing threat landscape.

GUI agents built on LVLMs are increasingly used to interact with websites. However, their exposure to open-world content makes them vulnerable to Environmental Injection Attacks (EIAs) that hijack agent behavior via webpage elements. Many recent studies assume the attacker to be a regular user who can only upload a single trigger image, which is more realistic than earlier assumptions of website-level administrative control. However, these works still fall short of realism: (1) the trigger's position and surrounding context remain largely fixed between training and testing, failing to capture the dynamic nature of real webpages and (2) the trigger often occupies an unrealistically large area, whereas real-world images are typically small. To better reflect real-world scenarios, we introduce a more realistic threat model where the attacker is a regular user and the trigger image is small and embedded within a dynamically changing environment. As a result, existing attacks prove largely ineffective under this threat model. To better expose the vulnerabilities of GUI agents, we propose Chameleon, an attack framework with two main novelties. The first is LLM-Driven Environment Simulation, which automatically generates diverse and high-fidelity webpage simulations. The second is Attention Black Hole, which transforms attention weights into explicit supervisory signals that guide the agent's focus toward the trigger region. We evaluate Chameleon on 6 realistic websites and 4 representative LVLM-powered GUI agents, where it significantly outperforms existing methods. Ablation studies confirm that both novelties are critical to performance. Our findings reveal underexplored vulnerabilities in modern GUI agents and establish a robust foundation for future research on defense in open-world GUI agent systems. The code is publicly available at https://github.com/zhangyitonggg/attack2gui.

Recently, the WebAssembly (or Wasm) technology has been rapidly evolving, with many runtimes actively under development, providing cross-platform secure sandboxes for Wasm modules to run as portable containers. Compared with Docker, which isolates applications at the operating system level, Wasm runtimes provide more security mechanisms, such as linear memory, type checking, and protected call stacks. Although Wasm is designed with security in mind and considered to be a more secure container runtime, various security challenges have arisen, and researchers have focused on the security of Wasm runtimes, such as discovering vulnerabilities or proposing new security mechanisms to achieve robust isolation. However, we have observed that the resource isolation is not well protected by the current Wasm runtimes, and attackers can exhaust the host's resources to interfere with the execution of other container instances by exploiting the WASI/WASIX interfaces. And the attack surface has not been well explored and measured. In this paper, we explore the resource isolation attack surface of Wasm runtimes systematically by proposing several static Wasm runtime analysis approaches. Based on the analysis results, we propose several exploitation strategies to break the resource isolation of Wasm runtimes. The experimental results show that malicious Wasm instances can not only consume large amounts of system resources on their own but also introduce high workloads into other components of the underlying operating system, leading to a substantial performance degradation of the whole system. In addition, the mitigation approaches have also been discussed.

Deep learning (DL) compilers are core infrastructure in modern DL systems, offering flexibility and scalability beyond vendor-specific libraries. This work uncovers a fundamental vulnerability in their design: can an official, unmodified compiler alter a model's semantics during compilation and introduce hidden backdoors? We study both adversarial and natural settings. In the adversarial case, we craft benign models where triggers have no effect pre-compilation but become effective backdoors after compilation. Tested on six models, three commercial compilers, and two hardware platforms, our attack yields 100% success on triggered inputs while preserving normal accuracy and remaining undetected by state-of-the-art detectors. The attack generalizes across compilers, hardware, and floating-point settings. In the natural setting, we analyze the top 100 HuggingFace models (including one with 220M+ downloads) and find natural triggers in 31 models. This shows that compilers can introduce risks even without adversarial manipulation. Our results reveal an overlooked threat: unmodified DL compilers can silently alter model semantics. To our knowledge, this is the first work to expose inherent security risks in DL compiler design, opening a new direction for secure and trustworthy ML.