CyberSec Research

Connected and Autonomous Vehicles (CAVs) rely on interconnected software and hardware components, including sensors, Electronic Control Units, in-vehicle infotainment systems, and telematics units, where vulnerabilities can compromise assets, users, and vehicle operations. These vulnerabilities are commonly documented as plain text in the Common Vulnerabilities and Exposures (CVE) database; however, security practitioners require structured information about affected assets, types of weaknesses, and attack behaviors to effectively mitigate the risks from these vulnerabilities. To this end, we evaluate open-weight Large Language Models (LLMs) for generating Structured Threat Information Expression (STIX), a well-known structured format for representing threat information, for CAV-related CVEs. We construct a dataset called CAV-STIXGen that maps CAV vulnerability descriptions to STIX domain objects (SDO), STIX relationship objects (SRO), Common Weakness Enumeration (CWE), and MITRE ATT&CK techniques mappings. Using this dataset, we evaluated 11 open-weight LLMs (4B to 120B parameters) across various prompting strategies and temperatures. Single-model configurations achieve F1 scores of 0.94 for SDO, 0.63 for SRO, and 0.99 for CWE mapping, while complete MITRE ATT&CK mapping remains challenging. In a multi-agent setup, Gemma-4-31B and Codestral-22B achieve F1 scores of 0.91 for SDOs and 0.43 for SROs, respectively. Lastly, we analyze CWE and MITRE ATT&CK co-occurrences to identify recurring threat patterns in the CAV domain, demonstrating how AI-assisted vulnerability-to-STIX translation can automate threat intelligence and prioritize defense in transportation security.

The 3rd Generation Partnership Project (3GPP)'s Fifth Generation New Radio (5G NR) is critical to supporting mission-critical services. However, 5G systems are vulnerable to smart jamming attacks that can propagate to applications running on top of these networks (i.e., cross-layer). The 5G gNB broadcasts resource scheduling information for the legitimate UEs over the air interface, with a prevailing assumption that this surface alone reveals nothing useful about a user device. However, we show that using the Downlink Control Information (DCI) is sufficient to degrade Application layer service quality, i.e., Denial of Service Quality (DoSQ), by inferring the Application layer Goodput (i.e., via side-channel analysis). Therefore, we present DoSQ, a protocol-aware attack that decodes per-slot DCI to inject interference onto the victim UE's Physical Resource Blocks (PRBs) within the same 1 ms slot, while a cross-layer classifier estimates the victim's Goodput state and trend from DCI features alone, without observing a single encrypted byte. Evaluated on a private 5G NR testbed against YouTube Live, DoSQ drives the target's Goodput down by up to 50% at sparse hit-rates, while a co-located non-target UE remains largely unaffected. Moreover, the classifier achieves a precision of 0.87 at the top 1% of attack-now confidence, a 4.21 times lift over the base rate. Furthermore, we propose an SSB frequency-time-hopping countermeasure that increases the attacker's resynchronization cost. The result is the first empirical measurement of a radio-to-application side channel that any protocol-aware adversary can exploit.

Gasp is a decentralized exchange designed as an application-specific Layer 2 (L2) rollup with omnichain connectivity, leveraging EigenLayer's restaked ETH for computation correctness and finalization. With a goal of being a consolidation layer for all crypto assets, the Gasp platform employs optimistic rollup technology to facilitate gas-free, native cross-chain swaps without reliance on traditional bridges, ensuring tokens retain their original L1 grade security. By combining an app-chain architecture with escape hatch mechanisms, Gasp guarantees withdrawal, while MEV minimization through Themis architecture reduces value extraction risks. Gasp's proof-of-liquidity framework unlocks staked liquidity, enhancing capital efficiency and liquidity depth by integrating staking with liquidity provisioning. Additionally, the protocol introduces a time-based reward mechanism, incentivizing long-term liquidity commitment via an asymptotic reward curve. This paper examines the current challenges in cross-chain communication, delineates Gasp's architectural innovations and security guarantees, and examines novel approaches to optimizing DeFi ecosystems.

Governments are increasingly mandating that LLM-generated content carry watermarks. The EU AI Act calls for markings that are "sufficiently reliable and robust." California's SB 942 requires disclosure that is "permanent or extraordinarily difficult to remove." Both mandates rest on an untested assumption: that watermark detection yields evidence reliable enough for courts. This paper tests that assumption directly. We evaluate three representative LLM watermarking methods -- KGW, Unigram, and the MarkLLM implementation of SynthID-Text -- against the Daubert admissibility criteria and the NIST SP 800-86 digital forensic process. To structure this evaluation, we propose a Forensic Readiness Score (FRS) framework with 12 criteria, three mandatory gates, and a 60-point scoring system. We focus on meaning-preserving paraphrase as the attack vector, since it is both legally realistic and difficult to dismiss as evidence tampering. The results raise serious evidentiary concerns. Out of 846 valid paraphrase runs across 15 diverse prompts per method, every single initially-detected KGW and Unigram text lost its watermark after paraphrasing -- 100% conditional removal. SynthID fared only slightly better at 98.3%. Even before any attack, false-negative rates were already high: 70% for KGW, 83% for Unigram, 80% for SynthID. The SynthID configuration also flagged 5.4% of paraphrased human-written controls as AI-generated and showed an 18.6% paradox rate, with 80% of its own pristine watermarked output landing in the uncertainty deadband. None of the three methods satisfy more than two of five Daubert factors. We also find that the FRS point-based scoring system, despite working as designed, cannot fully capture forensic uselessness -- a limitation worth noting for future framework design. These configurations, as tested, do not meet the evidentiary bar that courts require.

Safety defenses for large language models (LLMs) have been extensively studied, with existing approaches focusing on attack detection and refusal mechanisms. Such fixed-form direct refusal strategies may introduce the risk of prefix injection attacks. Recent work has explored a new direction that leverages humor as an indirect refusal mechanism to mitigate over-refusal in jailbreak scenarios and reduce prefix injection risks. However, this approach implicitly assumes that humorous responses are safe. Whether humorization itself introduces safety risks remains unexplored. To address this issue, we conduct an exploratory study involving over 30,000 real-world agent interaction records and 45 stand-up comedians, revealing practical safety concerns in LLM-based content humorization. Motivated by these findings, we propose \textsc{HumorSafe}, a novel framework for evaluating latent safety risk propagation during humorization. \textsc{HumorSafe} enables LLMs to learn harmful humorization patterns and use them to transform benign content into humorous content with safety risks. Across five frontier LLMs, we find that LLMs can introduce stereotypes and toxicity during humorization. We further propose \textsc{HumorPIA}, a prompt injection attack that exploits latent risks in humor-based defenses. \textsc{HumorPIA} preserves the appearance of safe humorous refusal while covertly injecting harmful content, allowing latent risks to evade existing detection mechanisms. Experiments show that it increases toxicity by 3.14$\times$ while maintaining an apparent safety rate of 97.8\% even under defense settings. Our findings highlight a gap in existing LLM safety evaluations under humorized settings.

The flourishing code hosting platforms and coding agents enable even beginners with private data to build tailored Machine Learning (ML) models using available code quickly. The training data for ML models, often regarded as private property (e.g., clinical records, transaction information), is at significant risk of information leakage. Property Inference Attacks (PIAs), as a significant type of privacy attack, aim to expose global property information of the training set. In this paper, we present Code-Poisoning Property Inference Attack (CPPIA), the first code-level PIA, which overcomes four limitations of existing works: insufficient attack performance, severe degradation of model accuracy, high computational overhead, and failure under defenses. We consider malicious code providers from code hosting platforms (GitHub) and coding agents (Codex). Upon downloading the poisoned code, data holders train models with their private data without professional auditing, subsequently releasing label-only APIs to the public. The adversary embeds the properties into secret samples during training and queries the trained model on these samples later to leak privacy. CPPIA offers 100\% attack accuracy without degrading model accuracy. It is also computationally lightweight and requires no shadow models. We evaluate the attack performance across four datasets, eight model architectures, eighteen properties, and under three defense mechanisms, demonstrating the universality and effectiveness of CPPIA.

Large Language Models (LLMs) are increasingly used for source code generation despite their outputs often exhibiting security vulnerabilities. Prior work shows that prompt engineering can mitigate such risks, yet (1) they focused on high-level prompting strategies, neglecting recent evidence that fine-grained syntactic variations can substantially alter model behavior; and (2) predominantly evaluate proprietary LLMs, limiting the applicability of their findings in industrial settings where self-hosted, open models are preferred for privacy, compliance, and deployment control. In this paper, we study how fine-grained syntactic constituents of prompts influence the security of open LLM-generated code. Using a parser-driven approach, we systematically generate syntactic variants of security-relevant code generation prompts and evaluate their impact on code security across multiple open LLMs and programming languages. Our results show that specific syntactic elements, such as constraints, guards, conditions, and concept bindings, and their position within the prompt consistently affect the likelihood of generating insecure code. These findings identify prompt syntax as a concrete security control surface and provide actionable guidance for reducing vulnerability risk in LLM-assisted development.

The integration of wireless communication technologies in industrial automation offers greater flexibility, but also exposes safety systems to a broader threat vector. Emerging regulations, such as the draft standard prEN 50742, mandate the convergence of functional safety and cybersecurity by requiring cryptographic security mechanisms directly in safety-critical communication. This paper presents an empirical evaluation of this safety-security convergence across a complete control chain, spanning from an IO-Link Wireless Safety device to a PLC via an OPC UA backbone. We measure the latencies and jitter of different Safety-Related Security Levels under prEN 50742 over Ethernet, Wi-Fi 6, and private 5G. Our results reveal that while cryptographic execution time is negligible, the resulting frame payload expansion severely restricts wireless fieldbus capacity, reducing the maximum number of devices per IO-Link Wireless track from 8 to 2. Furthermore, we demonstrate that, despite higher average latency, a private 5G provides sufficiently deterministic latency characteristics to preserve functional safety watchdog margins, unlike unlicensed Wi-Fi 6.

DICOM is the dominant protocol for exchanging medical images, yet many Internet-facing deployments lack basic security controls, exposing sensitive patient data to unauthorized access. Accurately measuring this exposure is complicated by honeypots, network telescopes, and other measurement artifacts that inflate published estimates. This paper presents a noise-aware study of Internet-facing DICOM services, combining active IPv4-wide scanning with passive honeypot deployments. We scan common DICOM ports using an ethics-constrained probe limited to association negotiation and C-ECHO. Then, we introduce a reproducible false-positive filtering method that identifies honeypots, telescopes, malformed responders, and other deception artifacts, reducing apparent exposure by 39%. After filtering, we identify 3,979 vulnerable DICOM deployments, all of which lack encryption and accept unauthenticated connections; 1,782 run outdated or deprecated software, and 1,551 carry known remotely exploitable vulnerabilities. Follow-up scans reveal that approximately 50% of exposed services show no evidence of maintenance over 5 months of observation, and that responsible disclosure led to only modest, short-term remediation. On the passive measurement, we deploy Dicompot and find that its raw session logs overstate activity by up to 83% due to generic TCP noise being incorrectly logged as DICOM sessions. After filtering, we observe a reconnaissance gap: Most actors issuing C-ECHO probes never escalate to data exfiltration, consistent with sophisticated actors fingerprinting the honeypot and disengaging before proceeding. Our results show that DICOM exposure measurements can be significantly distorted by deception and logging artifacts, but once corrected, reveal widespread risks to patient privacy and healthcare security that existing deception systems are insufficient to capture.

Designing hardware circuits resistant to side-channel attacks increasingly relies on simulation to predict device leakage before fabrication. Current functional verification simulators are designed for extended correctness-checking runs and are ill-suited for producing large numbers of short trace collections with slight input variants needed for side-channel analysis. We present Vogls: an open-source Verilog simulator built for side-channel analysis, that is the first simulator to combine compiled-code performance, full-timing simulation, and fine-grained control over the simulation state. Vogls simulates a timing annotated gate-level AES design 5.9 times faster than Icarus Verilog and is only 30% slower than Verilator on a PicoRV32 RTL design while also offering a more accurate timing model. Vogls provides a Python interface that allows simulations to be paused, forked, inspected, and mutated around points-of-interest without modifying the hardware design and thereby preserving timing and leakage characteristics. Furthermore, a trace-collection workflow runs setup code once and forks at the point-of-interest, eliminating the need to re-simulate from reset for every trace. A case study demonstrates Vogls on a differential power analysis attack that successfully recovers the key at RTL, GTL and full-timing GTL abstraction levels.

Cyber-attacks against exposed healthcare infrastructure threaten sensitive patient data and clinical operations, yet existing defensive tools for DICOM-based medical imaging systems provide limited interaction and are easily fingerprinted. We introduce DICOMHawk, a cyber-deception framework that emulates DICOM and PACS services using realistic interactions, dynamically populated medical records, and embedded honeytokens. In an 86-day comparison and a 347-day deployment across multiple networks, DICOMHawk attracted more valid sessions than Dicompot, avoided honeypot detection, and captured 49 medical-related attacks. The results show that realistic, long-term, multi-location deception improves visibility into threats targeting medical imaging systems.

Fully homomorphic encryption (FHE) schemes such as RNS-CKKS enable privacy-preserving services by allowing direct computation on encrypted data. While recent FHE compilers optimize FHE programs, they operate at the coarse-grained ciphertext level, where each ciphertext operation comprises a sequence of polynomial operations. At this granularity, the compilers miss optimization opportunities across ciphertext operations. This work presents Recifhe, a new multi-level compiler that supports not only ciphertext-level but also polynomial-level optimization. At the ciphertext level, Recifhe transforms a non-FHE input program into an FHE program by inserting ciphertext management operations and applies global optimizations. At the polynomial level, Recifhe eliminates redundant polynomial computations across ciphertext operations. Recifhe achieves a 1.25x speedup over ciphertext-level-only optimization.

With the rapid development of deep learning, its vulnerability has gradually emerged in recent years. This work focuses on backdoor attacks on speech recognition systems. We adopt sounds that are ordinary in nature or in our daily life as triggers for natural backdoor attacks. We conduct experiments on two datasets and three models to validate the performance of natural backdoor attacks and explore the effects of poisoning rate, trigger duration and blend ratio on the performance of natural backdoor attacks. Our results show that natural backdoor attacks have a high attack success rate without compromising model performance on benign samples, even with short or low-amplitude triggers. It requires only 5% of poisoned samples to achieve a near 100% attack success rate. In addition, the backdoor will be automatically activated by the corresponding sound in nature, which is not easy to be detected and will bring severer harm.

Backdoor attacks pose a critical threat to neural network models, allowing attackers to implant a backdoor during the training phase by manipulating a small portion of the training data. In security-sensitive applications such as voice interaction for autonomous driving, the presence of backdoor attacks introduces substantial security risks. This study focuses on implementing backdoor defense measures for speech recognition models in run-time, taking into account the characteristics of audio signals. We propose SpeechGuard, the first online backdoor defense pipeline designed to identify and purify poisoned audio samples. Specifically, we improve STRIP method to perform adaptive perturbation injection to detect and filter poisoned samples, named as S-STRIP. More importantly, we further consider the purification of poisoned samples. We utilize time-frequency (T-F) masking to suppress the expression of trigger signals and autonomously generate masks based on an autoencoder. The two-stage processing prevents the backdoor in the model from being triggered, and even input speech carrying triggers can be accurately predicted. Extensive experimental demonstrate that SpeechGuard can accurately filter out poisoned samples. Through purification, it can significantly mitigate the backdoor threat while maintaining a certain prediction accuracy.

Smart contract vulnerabilities pose severe financial risks, yet existing security tools largely stop at vulnerability detection, offering limited support for explaining whether reported flaws are exploitable, how attacks unfold, and what concrete damage they cause. To bridge this gap, we propose KASS (Knowledge-Augmented Attack Synthesis and Simulation), a multi-agent framework for executable smart contract exploit verification. KASS decomposes automated exploit generation into planning, generation, and testing stages, and integrates three complementary mechanisms: retrieval-augmented planning over real-world audit knowledge, formal generation and validation constraints that bind attack plans to executable proof-of-concept tests, and a hierarchical dual-loop refinement process that repairs code-level errors while triggering strategy-level replanning when attack assumptions fail. We evaluate KASS on 104 SmartBugs-Curated contracts across four vulnerability categories. Experimental results show that KASS successfully generates executable exploits for 94.23% of tested contracts; this rate is higher than previously reported results for REX and AdvSCanner on comparable SmartBugs-Curated subsets, and higher than our reproduced Claude Code baseline under the same evaluation protocol. On 11 real-world CVE-tagged contracts, KASS successfully validates 9 cases. Beyond exploit generation, KASS produces structured attack plans that document exploitation flows, quantify potential asset losses, and serve as semantic false positive filters for static analysis tools.

Multimodal AI agents increasingly rely on persistent long-term memory to ground generation in past visual and textual episodes. We show that unconditional trust in visual data creates a critical vulnerability. We propose Lucid, a black-box adversarial framework that compromises multimodal memory pipelines under a strictly image-bounded threat model, requiring no access to the target MLLM, target retrieval encoder, or the text channel. Lucid crafts imperceptible perturbations to enable two distinct failure modes based on the availability of historical context: (1) Memory poisoning, an in-context attack where the adversarial image replaces a benign one whose content is reinforced by prior textual context, reliably corrupting visual recall and steering the agent toward attacker-chosen narratives; (2) Memory injection, an out-of-context attack where the adversarial image replaces a benign one in a conversation turn devoid of prior textual grounding, causing the agent to generate attacker-influenced responses with no corrective signal from memory. We evaluate Lucid across various conversation domains and five black-box memory architectures, including graph-structured, LLM-summarized, and commercially deployed systems. Lucid achieves 61.6% ASR on poisoning and 58.4% ASR on injection, exposing a structural vulnerability in multimodal memory pipelines.

The rapid adoption of artificial intelligence (AI)-driven and agentic workflows is transforming traditional government and enterprise systems into language-based, tool-using and increasingly autonomous infrastructures. Conventional key management services authenticate who may invoke a cryptographic primitive, but remain agnostic to which workflow steps are authorized at runtime: an authenticated agent can still be hijacked by direct or indirect prompt injection into executing malicious actions that satisfy identity checks yet violate user intent. We present \emph{Neural Cryptographic Services} (NCS), an active security governance plane grounded in a neural-symbolic design and interposed between LLM agents and privileged tools. Under NCS, an untrusted neural planner compiles natural-language instructions into structured plan drafts but holds no execution authority. Execution is instead gated by a deterministic symbolic controller operating over an offline-signed, hash-chained instruction stream. NCS verifies the signature, incrementally validates the hash chain, releases exactly one instruction payload at a time, and enforces strict binding between the agent's proposed tool parameters and the verified payload. Mismatched or out-of-order tool calls are rejected fail-closed, while previously verified states are retained for post-hoc auditing. We evaluate NCS on AgentDojo and a custom argument-hijacking benchmark. NCS drives attack success rates to near zero while preserving acceptable utility on benign workflows. NCS thus reframes agent security from asking whether a model's intent complied to asking whether a proposed dispatch matches a cryptographically authorized step.

Following Goldwasser, Rothblum, Shafer, and Yehudayoff, who defined a framework for interactive proofs of learning [ITCS'21], we initiate the study of non-interactive proofs of learning. We define and study a new notion: Publicly-Verifiable Certificates of Statistical Validity (pvCSVs), which allow for public, distributionally-robust certification that the result of a learning algorithm is valid. In a pvCSV, a learner publishes a hypothesis $h$ and corresponding certificate $π$; then, any user, who holds a user-specific distribution, can read the pair $(h,π)$ and determine efficiently whether the hypothesis is valid according to the user-specific distribution. We construct pvCSVs in the context of Adaptive Statistical Query (SQ) Algorithms. To certify SQ algorithms that makes $k$ adaptive queries, we construct pvCSVs where the sample complexity scales with $O(\log k)$, whereas the sample complexity of the best learning algorithms scale with $\tilde{O}(\sqrt{k})$. More generally, we study proof systems for learning in the SQ model, demonstrating the model's strengths as well as its limitations.