Papers

Evaluating the ability of large language models (LLMs) to discover software bugs is increasingly important. Existing benchmarks typically evaluate this capability by asking the model to generate a proof-of-concept input that triggers a predefined target vulnerability. However, this setup may overlook valid crashes discovered by the model when they do not match the predefined target. As a result, the evaluation may not reflect the model's real capability. We present FuzzingBrain-Bench, a benchmark for assessing AI models' ability to discover bugs in open-source software. Models are given an open-source project and a sanitizer-instrumented harness in a self-contained Docker image. Their goal is to generate inputs that trigger as many distinct crashes as possible through the harness. A model's performance on each challenge is scored based on the number of distinct crash signatures it produces, capped at a predefined maximum and weighted by a difficulty coefficient. FuzzingBrain-Bench V1 consists of 77 challenges drawn from 43 open-source projects, with 36 C, 32 C++, and 9 Java/JVM challenges. We evaluate Claude Haiku 4.5, Claude Sonnet 4.6, and Claude Opus 4.8 on the full benchmark. Claude Opus 4.8 performs best, triggering crashes in 60 of 77 challenges and achieving a score of 196 out of 579. None of the three models triggers a crash in 13 challenges. The FuzzingBrain-Bench corpus and harnesses are publicly available at https://github.com/fuzzingbrain/FuzzingBrain-Bench.

We describe and analyze a protocol for mutual co-signing of a document by two mobile devices that (i) communicate only over a one-way, lossy, low-bandwidth optical channel (an animated on-screen code read by the counterparty's camera), (ii) use no intermediary server on the trust path, and (iii) use no certificate authority. Trust in each party's public key is instead grounded in a hardware attestation token produced by the platform secure element, transported in full over the visual channel by a rateless (fountain) code and cryptographically bound into the co-signature. The core technical contribution is a two-stage hash anchor that removes the circular signing dependency inherent to interactive co-signing: the first party commits to the document before the identity of the second party is known, and the second party's identity is later bound to that commitment without invalidating the first signature. We give a threat model, define four security properties (anchor binding, co-signature inseparability, attestation-bound key provenance, and post-signing tamper evidence) and reduce them to standard assumptions (collision resistance of H and EUF-CMA security of the underlying signature scheme), with the secure element modeled as an ideal signing oracle. We report a working instantiation on iOS/Android using ECDSA P-256 in the Secure Enclave/StrongBox, SHA-256, Apple App Attest / Play Integrity, and an LT-style fountain code, together with an independent third-party verifier that recomputes all anchors and checks both signatures fully offline.

The development of federated learning (FL) techniques has helped improve the privacy preservation of users' data and extended the applications of machine learning models. However, the involvement of a large number of users in FL also creates open opportunities for different adversaries, such as poisoning attacks, Byzantine attacks, and adversarial example attacks. Yet, recent research has disclosed that existing poisoning attacks and Byzantine attacks can not achieve satisfactory penetration in realistic FL scenarios caused by strong assumptions, e.g., client selection rate, and the ratio of malicious attackers. In this paper, the transferability of adversarial examples among different client models is analyzed to understand the relation between adversarial examples and clients' data distribution. Moreover, to mitigate the attacks of transferable adversarial examples, we design a defense mechanism stemming from the transferability of model robustness by adversarial training. As a result, through theoretical analysis of transferability, we gain insights into adversarial examples and the vulnerability of federated learning systems. Our proposed adversarial attack and defense methods are evaluated via real-life datasets in various settings to show their performance over the existing state-of-the-art methods.

Identifying which post-quantum algorithm a compiled binary implements is a core problem in binary analysis, and it becomes acute once the binary has been stripped, statically linked, and optimised, since the symbols, library dependencies, and runtime behaviour that conventional discovery tools rely on are then gone. This paper presents Kestrel, a static analysis method that identifies an algorithm from the number-theoretic transform constant tables its arithmetic depends on, and applies it to the standardised lattice schemes ML-KEM and ML-DSA, for which no prior method could confirm, from a shipped binary alone, that a quantum-vulnerable algorithm had been replaced by its approved successor. The fingerprints Kestrel derives from public scheme parameters are localised by a normalisation-and-multiset-matching procedure; the false-positive probability is established analytically. Across four independent implementation lineages and all build transformations, including compiler-level obfuscation, Kestrel achieved recall of 128 of 128 with zero false positives. Applied to 6,224 binaries on a production Linux system, it disclosed twelve uncatalogued programs containing ML-KEM, among them the OpenSSH key-exchange program and the container-management stack, where post-quantum code had entered production through the language runtime without the awareness of the projects distributing them. Kestrel distinguishes genuine implementations from advertised claims not backed by the underlying code, attributes each detection to its originating codebase, and, in a forensic disk-image trial, recovered a detection from unallocated space after the deleted binary could no longer be reconstructed. It thus provides a practical basis for cryptographic migration assurance, compliance verification, software supply-chain inspection, and post-quantum forensic examination.

Self-evolving Skill harnesses (AutoSkills, Hermes Agent) generate more advisory orchestration automatically; their reported gains are efficiency, not safety. This misses the actual gap: a Skill describes how an agent should behave; a Policy decides which behavior is allowed to become an action. Today's format covers the first with markdown and scripts; the second is left to the model. Generating more Skills scales the gap, not the safety, especially when a wrong invocation can unlock a door or move money. Two adjacent attacks are documented: malicious skills compromising cloud software, and jailbroken LLM-controlled robots causing physical harm. Their intersection, malicious agent skills causing physical harm, follows directly but has not been reported. We name this class Borrowed Authority: Skills format gives the receiving agent no typed way to reject an inter-agent permission claim, so a malicious or misused Skill can drive actuation by attaching one. We propose Edge Skillguard, a typed authority layer that lives inside the Skill artifact rather than between tools as workflow engines do, with guards over world state and sensor evidence. On a live edge control-plane testbed, the guards reject 60/60 borrowed-authority requests across five attack variants without blocking benign requests, and the result holds at 5x scale and across hosts over a Tailscale mesh. These results suggest that high-risk Skills should co-package typed invocation policy with procedural knowledge, so that physical actions depend on machine-checkable evidence rather than peer-agent claims.

Retrieval-Augmented Generation (RAG) enhances large language models by grounding outputs in external knowledge, improving factuality and reducing hallucinations. At the same time, the retrieval-augmented pipeline introduces new robustness and security risks, including corpus poisoning, backdoor attacks, privacy leakage, and fairness violations. Despite rapid progress in this area, existing surveys remain limited in their treatment of attacker objectives, threat models, and stage-specific defenses across the full RAG pipeline. This survey presents a unified and pipeline-aware overview of RAG robustness. We formalize threat models over the corpus, retriever, and generator, and organize attacks into three main objectives: accuracy, privacy, and fairness. We further review defenses from a pipeline-aware perspective, covering the retrieval, rerank, generation, and traceback stages. In addition, we summarize robustness benchmarks and explainability methods for more deeply evaluating and explaining RAG robustness.

AI coding assistants generate Infrastructure as Code, yet no work has examined whether this code meets security requirements. This matters because security smells in infrastructure code propagate to deployed systems, producing infrastructure that is insecure and untrustworthy. We evaluate 16 AI models generating Ansible roles for Apache Tomcat v10 and MongoDB v7, analysing 278 Ansible roles against CIS benchmarks. Without security guidance, all 16 AI models produced code containing security smells, resulting in vulnerable infrastructure that fails compliance verification and underperforms code written by human developers. We introduce an approach integrating Ansible best practices and CIS benchmarks into prompts through an extended CO-STAR framework, enabling security smell prevention during synthesis rather than detection after deployment. When this approach is applied, 4 out of 16 models generate compliant code, with the leading model achieving 95%-100% CIS compliance, a fourfold improvement over humans at 23%-43%, with overall code quality improving by 19%-49%. The remaining 12 models fail not because they cannot generate code but because they cannot follow instructions with multiple constraints. For capable models, the approach requires no retraining and can be adopted through system prompts.

LLM agents routinely include privacy-sensitive data (PSD) in tool call arguments beyond what the invoked tools require, crossing trust boundaries to third-party services on every invocation. A controlled measurement on three production LLMs (GPT-4o, Claude 3.5 Sonnet, Llama-3.3-70B) shows that 81–88% of tool calls include unnecessary PSD under default prompts; explicit privacy instructions still leave 36–76% over-sharing. Existing defenses gate calls (allow/block) or label flows (information-flow control) but cannot rewrite argument values, and PII detection tools miss implicit PSD like “Memorial Sloan Kettering” (a hospital name that implies a diagnosis). We present \system{}, a middleware that intercepts tool calls and rewrites their arguments to the minimum data necessary for tool functionality, combining schema-aware necessity analysis with four operations: removal, generalization, substitution, and truncation. Live validation on 307 tool calls across the three LLMs above reduces privacy cost by 81.2–92.0% at 100% argument-level task validity (TOST equivalence $p{<}0.001$ at $\Delta{=}1.0$); on 25 unannotated Model Context Protocol (MCP) schemas, by 79.0% with no minimum\_necessary metadata. An optional LLM content-necessity layer strips task-irrelevant PSD from otherwise-necessary free-text fields, raising live-LLM reduction to 85.1–95.6% and author-schema reduction from 71.1% to 90.9%. Median latency is 1.77\,ms.

Decompilation has become a foundational technique in software engineering and security analysis, and it is now advancing through the integration of modern machine learning (ML) approaches. This article presents a systematic review of decompilation studies published over the past decades and develops a comprehensive taxonomy of methodologies employed in contemporary research. We further examine trends in evaluation metrics, tools, and benchmarks used to assess state-of-the-art approaches. Our review reveals key challenges, such as the lack of reliable ground truth and the absence of standardized benchmarks, which hinder rigorous comparison. Finally, we outline future research directions.

Large Language Models (LLMs) increasingly generate code from natural-language prompts, making prompt engineering a key mechanism for shaping the security of generated software. Structured and security-oriented prompts are widely used to encourage safer code, yet their effects extend beyond whether detected weaknesses are simply present or absent. Using 424 security-sensitive Python tasks, we generate solutions with GPT-4o and LLaMA 3.1-8B under five prompt variants that progressively add structural and security guidance, and evaluate them with Bandit and CodeQL along two axes: generation compliance and security weakness prevalence, severity, and CWE distributions. Structured prompting substantially reduces refusals (e.g., GPT-4o invalid outputs drop from 338 of 424 to 37-52), enabling large-scale analysis, but security-oriented refinements do not consistently reduce overall weakness prevalence. For GPT-4o, stronger prompts primarily redistribute risk: high-severity findings fall (20.8% to 13.6%) while low-severity findings rise (32% to 43.5%); LLaMA shows weaker, less consistent shifts. We also observe security-driven semantic drift, where stricter prompts silently remove or rewrite explicitly requested unsafe constructs. Overall, prompt structure improves compliance but is an unreliable substitute for robust security controls in LLM-assisted development.

Enterprise cybersecurity research draws on a wider range of methods than any single community routinely teaches. Researchers face a selection problem before they face a technical one: a study may simultaneously need a systematic review, a design-science artifact, a controlled detection experiment, an interview study, or an attack-graph model. This paper addresses that problem in two ways. First, it provides a narrative review and synthesis of methodological practices across a verified corpus of 151 works. We organise these practices into eleven methodology families, detailing for each what questions it answers, the strength of its supporting evidence, and its common failure modes. Second, we convert each family into an executable protocol comprising ordered steps, required instruments, evaluation criteria, common validity threats, and a reporting checklist. Every protocol is also visually mapped to make the sequence, decisions, and threats legible at a glance. We also treat contradictions in the literature as evidence. For example, reported rankings of intrusion-detection algorithms are wildly inconsistent across individually careful studies. We argue this pattern is most parsimoniously explained by variations in evaluation design rather than the algorithms themselves, as these studies differ in design dimensions known to shift results by more than the margins separating the algorithms. Ultimately, the evidence supports methodological pluralism disciplined by explicit validity reasoning. We conclude that researchers must match their evaluation design to the decision under study, triangulate technical against organisational evidence, explicitly state the population a result generalises to, and report the conditions under which the result would not hold.

In a recent breakthrough, Yamakawa and Zhandry (J. ACM 2024) constructed a proof of quantumness in the quantum random oracle model (QROM) in which the quantum prover samples a codeword preimage of a publicly computable function H. They conjectured that given any H, a successful prover must sample their preimage from a high-entropy distribution over possible answers. If true, this would give a certifiable randomness protocol in the quantum random oracle model. As partial evidence for their conjecture, Yamakawa and Zhandry proved the security of their certifiable randomness protocol assuming the Aaronson-Ambainis conjecture. We prove the security of the certifiable randomness protocol of Yamakawa-Zhandry unconditionally, without relying on the unproven Aaronson-Ambainis conjecture, against low query-depth quantum adversaries: specifically, adversaries that make up to o(\log \lambda) adaptive quantum queries to the random oracle.

We introduce masked auto-correlation, a new primitive for the cryptanalysis of symmetric-key primitives, together with a quantum attack pipeline built on it. For a permutation $f$, output masks $\alpha,\beta$, and an input difference $w$, masked auto-correlation (MAC) measures the correlation between the masked outputs $\alpha\cdot f(x)$ and $\beta\cdot f(x\oplus w)$. The associated masked differential-linear (MDL) approximations strictly generalize several classical techniques; ordinary linear cryptanalysis, differential-linear cryptanalysis, and the differential-linear connectivity table all arise as special cases. Our central object of study is the problem of finding mask pairs with large masked cross-correlation – those that yield powerful distinguishers – which we call MAC Fishing. We give a constant-query quantum algorithm that samples such pairs according to their squared correlation, and we prove an exponential classical lower bound of $\Omega(N/\log N)$ queries, by adapting the hardness of Fourier Fishing. To our knowledge this is the first result pairing a quantum upper bound with a classical lower bound for the core task of identifying high-correlation approximations, making quantum algorithms an absolute necessity. Building on this, we analyse the distribution of masked auto-correlation for random permutations, and then construct capacity-based distinguishers and key-recovery attacks, both classically and with a quadratic quantum speed-up using amplitude estimation. We validate our claims with experiments on reduced-round mini-AES.

Cybersecurity education is facing more challenges as AI-driven attacks are becoming increasingly realistic and difficult to detect. Traditional video-based cybersecurity training in higher education often suffers from both low engagement and limited effectiveness. This dilemma motivates educators to explore innovative approaches, such as AI-powered gamification, which can deliver engaging, meaningful, and personalized learning experiences. By presenting content in a more interactive and user-friendly way, these methods have the potential to significantly improve both learner engagement and educational outcomes. This paper explores AI-powered gamification in cybersecurity education through the development of several short, mobile-friendly games. These games cover a range of topics from password security to text and phone scam recognition, incorporate multiple gamification strategies, including quiz-based, narrative-based, and simulation-based designs, as well as interactive formats such as TikTok Mini-Games. We conducted a two-tiered evaluation with 59 college students (comprising 9 technical experts and 50 general users), and the results indicate the potential of AI-powered gamification to improve engagement and increase attention to cybersecurity topics in higher education.

LLM-based agents can interact with external environments through tool invocation, but this capability also introduces security risks such as file modification, information leakage, and unauthorized actions. Existing guardrails often evaluate completed trajectories, leaving pre-execution monitoring of step-level actions underexplored. We propose StepGuard, a step-level guard model that can audit completed agent trajectories and check tool actions before they are executed. To train StepGuard, we introduce StepGen, an automatic data engine that generates safe and unsafe trajectories with the same context but different actions at the risky step. To further reduce over-defense and under-defense, we propose Balance-GRPO, which dynamically balances learning between safe and unsafe actions based on their observed accuracy. Experiments show that StepGuard achieves the highest average accuracy among open-weight guard models, with performance comparable to GPT-5.4. When used to guard agents on AgentDojo and AgentDyn, StepGuard reduces mean attack success rate by 77.3% relative to the no-guard setting, while mean utility drops by only 2.8 percentage points.

Intrusion detection research in Industrial Control Systems (ICS) heavily depends on public datasets, yet no prior work has systematically assessed whether the collective dataset corpus supports current evaluation claims. This paper addresses this gap through a meta-review of 18 studies between 2019 and 2026, from which 83 ICS, or ICS directly related, cybersecurity datasets are identified, harmonised, and characterised using a unified five-dimensional taxonomy. The taxonomy reveals that the corpus is structurally skewed: 85.5% of datasets concentrate on late-stage OT Disruption tactics, cross-stage IT/OT progression sequences are present in only 8.4% of cases, field-device evidence at Level 0 of the Purdue hierarchy is effectively absent, and operationally sourced data accounts for only 15.7% of the collection. A parallel audit of evaluation practices shows that zero report streaming evaluation, fewer than half apply disciplined train/test partitioning, and only two satisfy reproducibility requirements. Furthermore, a taxonomy-evaluation coupling analysis shows that dataset imbalances constrain the scope and feasibility of several evaluation practices. Based on these findings, we identify three structural imbalances: i) architectural shallowness, ii) progression compression, and iii) cross-domain substitution, and derive a coordinated research agenda which covers cross-stage corpus construction, temporally structured benchmarking, event-level label standards, and governance frameworks for operational data sharing.

As mobile phone adoption has surged, so have scams involving these devices. One such scam, known as SMiShing (or smishing) after Short Message Service (SMS), involves fraudsters sending phishing links via mobile texts. Despite the prevalence of SMiShing, there is a lack of data on who is most vulnerable to these attacks. Prior research on phishing (its email counterpart) suggests that susceptibility may vary by demographic and contextual factors. In two large-scale surveys, we use a previously published simulation method to collect data from representative samples of U.S. adult mobile phone users. Our findings indicate that younger individuals and college students are particularly vulnerable. Participants struggled to correctly identify legitimate messages, with the second study providing comparisons of financial message variants. Researchers, regulators, and telecoms can help users by creating mobile-specific interventions for under-24 and university customers and adding verifications and warnings.

The primary goal of a distributed storage system is to ensure that clients can both write and read data in a reliable and consistent manner, even in the presence of failures. While existing asynchronous verifiable information dispersal (AVID) protocols achieve optimal space complexity for storage and communication complexity for data retrieval in a Byzantine setting, the crucial operations of data dispersal and node recovery have received less attention. We propose an efficient AVID protocol that simultaneously guarantees low complexities for dispersal, storage, retrieval, and recovery. At the core of the proposed protocol lies a novel mechanism to encode data in a two-dimensional matrix and a bespoke dispersal algorithm. The protocol maintains an optimal communication complexity for retrieval while substantially improving upon the state of the art for recovery. Additionally, we describe how the protocol can be modified in order to reduce the space complexity and the communication complexity for dispersal, at the expense of a higher communication complexity for retrieval and recovery. As the proposed protocols strike a balance across all considered metrics, they are suitable for a broad range of real-world use cases.

Constructing syntactically correct and cryptographically valid message sequences is essential for protocol state machine learning, conformance testing, and fuzzing. Unlike plaintext protocols, cryptographic protocols involve complex cross-message state dependencies and cryptographic computation constraints. Existing automated approaches predominantly target text-based or plaintext protocols, leaving cryptographic message construction largely manual. We present SeriCrypt, an LLM-driven, context-aware serialization framework for cryptographic protocols. It employs a large language model to extract field constraints, state dependencies, and cryptographic computation rules from unstructured protocol specifications into a unified structured intermediate representation, formally characterized by a domain-specific language for cryptographic protocols (CDSL). A protocol-agnostic execution engine parses CDSL declarations, automating field value resolution, cryptographic primitive invocation, and byte-stream serialization. As case studies in protocol security testing, we use the framework to construct violation messages targeting specification-defined security constraints and to support protocol fuzzing, evaluating it on mainstream implementations of TLS 1.2/1.3, IKEv1/v2, SSH, and TLCP. SeriCrypt generated message sequences accepted by all evaluated implementations and completed handshakes in every scenario. Security constraint testing revealed five specification violations, and fuzzing reached deeper protocol states with higher code coverage than mainstream fuzzers under the same time budget, demonstrating the framework's practical value for cryptographic protocol security testing.

System prompts can be extracted from commercial LLMs with over 80% success and redeployed at zero cost, yet a prompt owner has no way to verify whether a suspected deployment is a clone. We propose Black-Box Behavioral Fingerprinting (BBF): the prompt owner registers a behavioral signature from model outputs and later tests whether a suspect deployment matches that signature more closely than an unrelated baseline. BBF requires only black-box API access. Through a large-scale study (4 model families, 8 benchmarks, 288{,}000 responses), we find that prompt choice explains 24.4% of output variance and same-model detection reaches AUC 0.876. Cross-model performance is bounded by detector identity, with off-diagonal AUC ranging from 0.845 (Claude as detector) down to 0.665 (Qwen) and overall mean 0.725. BBF resists non-adaptive prompt paraphrasing (AUC $\geq 0.889$) and is robust to imperfect extraction, but a single-sentence formal-tone prefix can collapse detection on short structured outputs (MNLI 0.978 $\to$ 0.547), isolating style-invariant detection as the key open problem. Diagnostic Query Optimization, a zero-cost query selection rule, adds $+0.120$ to cross-model AUC.

Graph Neural Networks (GNNs) represent a promising solution for Machine Learning (ML) based Network Intrusion Detection Systems (NIDS), thanks to their ability to leverage both network flow features and topological patterns. While GNN classifiers demonstrate superior robustness against feature-based adversarial attacks compared to other ML detectors, they remain vulnerable to structural adversarial attacks, where an attacker perturbs the underlying network graph topology by injecting edges or inserting nodes. Such attacks pose a realistic and severe threat, undermining the reliability of GNN-based NIDS in practical deployments. While countermeasures have been proposed in the literature, they often rely on assumptions that are unrealistic in real-world cybersecurity scenarios. In this paper, we propose a defense framework based on adversarial training to strengthen GNN-based NIDS against structural attacks. We generate adversarial samples by strategically replacing the source and destination nodes in benign network flows, thereby efficiently mimicking edge injection attacks. We evaluate our approach on two widely used datasets (CTU-13 and TON-IoT) using E-GraphSAGE as the base GNN classifier. Experimental results show that our approach produces hardened detectors with superior detection performance on clean graphs and enhanced robustness against structural adversarial attacks.

The transition to post-quantum cryptography is reshaping the Secure Shell (SSH) protocol for remote administration. Post-quantum key exchange has been deployed in OpenSSH and is being standardized, while SSH authentication largely remains a signature-replacement effort. This path preserves the familiar public-key credential model, but inherits the size and computation overhead of post-quantum signatures, which can increase latency, traffic, and server-side load. KEM-based authentication offers a natural alternative to this signature-centric path, and SSH makes this especially attractive at the user-authentication layer, which is method-extensible, separated from transport-layer key exchange and host-key authentication, and already protected by the established channel. We present a drop-in KEM-based user-authentication method for SSH that replaces client public-key signatures with a session-bound challenge-response proof. The method fits into SSH's existing user-authentication framework, preserving the public-key credential model and enabling incremental deployment alongside existing methods. We provide a reduction-based security argument in the post-quantum ACCE framework, implement the design in OpenSSH using liboqs, and evaluate it under representative RTTs, TCP initial-window settings, and post-quantum migration configurations. Our results show that KEM-based authentication is competitive with compact signature-based authentication under representative network settings, while reducing median handshake latency by up to about 10% against large-signature hybrid baselines. The advantages are clearer when post-quantum signatures stress transmission or computation: median latency under small TCP initial windows falls by up to 7.3% versus ML-DSA and 17.9% versus SLH-DSA, while server-side online cryptographic cost is 59.1% lower than that for ML-DSA in the same NIST category.

MLLMs are increasingly deployed in user-facing applications, yet they inherit backdoor risks from the pipelines used to construct them: triggers may reside in images, texts, or both. Existing model-level backdoor removal methods, largely designed for conventional classifiers, show limited effectiveness on MLLMs, while MLLM-specific defenses mainly operate at inference time, filtering suspicious inputs without removing the backdoor embedded in the model. To address this gap and eliminate latent backdoors from MLLMs at their source, we present RACER, a model-level repair framework motivated by a key observation: backdoors induce abnormal layer-to-layer evolution in internal representations, which we term the layer-wise inconsistency anomaly. Importantly, this anomaly is modality-dependent, concentrating primarily in the token region encoding the trigger features that the backdoor model actually relies on. RACER therefore decomposes the fused representation into visual and textual token regions, normalizes their layer-wise inconsistency separately, and recomposes them using modality-aware weights over a deep-layer window, yielding a region-aware inconsistency objective that better captures localized backdoor-induced anomalies. Through a min-max optimization, this objective drives worst-case perturbation synthesis and adversarial fine-tuning against the resulting perturbation to repair the model, suppressing the deep representational directional shifts on which backdoor behaviors rely. RACER requires only 100 clean samples and no knowledge of the trigger, attack objective, or even whether the input model contains a backdoor. Evaluations on three open-source MLLMs across 36 backdoor settings spanning image, text, and multimodal triggers show that RACER reduces the average ASR to 1.1%, reaching 0% in 32 settings, while preserving clean-task utility on both backdoor and clean models.

Taint analysis is a fundamental technique for detecting sensitive data leaks in Android apps. However, traditional static tools, such as FlowDroid, still face well-known challenges due to the complexity of accurately modeling the Android framework. In this paper, we investigate whether off-the-shelf Large Language Models (LLMs) can effectively reason about taint flows in Android apps. Our preliminary approach relies on an agentic interaction strategy, enabling the LLM to iteratively explore code and reason about data flows. We conduct an initial evaluation on the DroidBench benchmark against FlowDroid, where our approach outperforms the baseline: Gemini-3 Flash achieves an F1-score of 0.96, compared to 0.55 for FlowDroid. In particular, we observe improvements in challenging categories such as inter-component communication (0.95 vs. 0.17), implicit flows (0.94 vs. 0.00), and reflection (1.00 vs. 0.50), where FlowDroid typically struggles. On a small set of real-world apps, the LLM-based approach also identifies additional potential data leaks not reported by FlowDroid. These preliminary findings suggest that LLM reasoning may effectively complement traditional static taint analysis, motivating future research on hybrid LLM-enhanced taint analysis pipelines.

Telephone fraud is pervasive and costly, but its inner workings are rarely observed at scale. We analyze a complete corpus of 10,211 inbound scam and spam calls – 913 hours of audio and 330,956 transcribed turns from 5,780 distinct numbers – collected over 54 days by an AI voice-agent honeypot that answered callers and kept them talking, and introduced in a companion data descriptor. We separate outright scams, which solicit sensitive information, from the larger stream of predatory but legal lead generation ("spam") that feeds them. Scam operations keep office hours (6.6x more calls per weekday than weekend day); thousands of disposable numbers run a small catalog of recycled scripts (thirty opening clusters, half the traffic in the top five); and callers solicit identity anchors – a home address and a date of birth – far more often than payment credentials, pressing through persistence and manufactured authority rather than overt threats. Our central experiment asks: does it matter who picks up? Every seeded lead carried one of ten fictitious identities drawn uniformly at random, so the identity a fraud operation reaches is fixed before the caller exists. Across 1,823 randomized calls, scammers spent about 15% more conversational turns per decade of the target's apparent age (rate ratio 1.15, 95% CI 1.08-1.23; randomization p = 0.005) – yet what they asked for did not change (26.3% of calls reached a request for sensitive information; odds ratio 0.99 per decade, 95% CI 0.90-1.08). A second experiment casts early detection as a benchmark: from a scammer's opening lines alone, on a caller-disjoint split, escalation is predictable at 0.72 ROC-AUC from the first line and 0.87 by the eighth, and a plain bag-of-words classifier matches a fine-tuned on-device language model. Telephone fraud emerges as a templated industry that varies how hard it works a target, but not what it wants.

LLM-based multi-agent trading systems, in which specialized agents collaborate through structured communication to produce trading decisions, are moving rapidly from research prototypes to live deployments that control real assets. The same inter-agent communication that makes them effective also exposes them: a corrupted signal can propagate to the final decision and translate into realized financial loss. Unlike prior attacks that presume privileged access to system internals, we restrict the adversary to what is practically reachable—the source data and prompts agents consume—yielding a low-barrier, and thus democratized threat model instantiated as role-specific adversaries. We present the first systematic empirical study in the financial domain to characterize how an adversarial signal enters a multi-agent trading system and how far it survives toward the decision. Along the role axis, we decompose a widely-used trading pipeline into four functional roles—Analyst, Researcher, Trader, and Risk Manager—and pair each with an attack matched to its interface. Along the structural axis, we evaluate four communication topologies under data- and agent-level attacks, using the Adversarial Signal Preservation Score (APS) as a post-hoc lens on why some designs are more robust than others. We conduct experiments across five assets, two backbones, and two target directions. A central finding is that no architecture is inherently robust. These findings provide insights for the future design of safer and more robust agentic trading systems.

LLM agents integrated with external resources gain complex task capabilities, yet the unified natural-language context channel makes them vulnerable to injection attacks: untrusted external data may be dynamically parsed as behavior-guiding instructions during LLM inference, thereby subverting the agent's decision. Existing defenses focus on static detection or isolation of malicious content at the input/output level, remains insufficient for detecting such dynamic inducements that arise during model reasoning. We propose Attnlocate, a runtime framework for fine-grained localization of context spans that genuinely influence tool-calling decisions, i.e., behavior-guiding instructions. Attnlocate casts this localization problem as an object detection task, aiming to detect the distinctive activation traces induced by behavior-guiding instructions within the attention matrix. Specifically, we design a multi-head, multi-layer attention aggregation scheme to construct a token-level feature space tailored for object detection. Then, a 1-D U-Net equipped with an anchor-free detection head is deployed to detect these spans. Finally, based on the authority of the provider from which the detected behavior-guiding spans originate, Attnlocate dynamically adjudicates malicious invocation attempts. We evaluate Attnlocate across ten agent configurations from five LLM families, covering scenarios involving indirect prompt injection and tool poisoning. Attnlocate achieves a mean IoU of 0.743, an average AUROC of 0.956, and a 0.934 true-positive rate at 0.067 false-positive rate. It also transfers effectively across unseen models and supports authority policy adaptation without retraining.

The emerging W3C WebMCP proposal enables LLM agents to invoke tools exposed by web pages. In multi-party web environments, however, integrating agent execution into a browser security model centered on the Same-Origin Policy (SOP) leaves insufficient provenance and lifecycle guarantees for agent-accessible tools, creating three risks: subject-attribution spoofing, uncontrolled tool lifecycles, and semantic prompt injection. We propose WebMCP-Phalanx, a dual-layer agent runtime architecture. Its first layer provides a browser-native trust anchor that binds each tool to its registering principal through cryptographically protected capability credentials and propagates provenance labels throughout the tool lifecycle. Its second layer separates semantic inspection from privileged tool use. A Quarantine Agent (Q-LLM), without tool invocation authority, inspects tool metadata, outputs, and page-supplied content for prompt injection. Validated content is then forwarded to a Privileged Agent (P-LLM) for execution, while the Q-LLM's internal state remains hidden from page scripts. Empirical evaluation shows that the browser-native ownership mechanism reduces revocation and overwrite attack success from 100% to 0%. The dual-agent runtime blocks all 80 prompt-injection attempts embedded in tool descriptions and limits tool-return attacks to 2 successful cases out of 80. Across experiments, task utility remains statistically indistinguishable from the no-attack baseline. Under a white-box adaptive attacker, however, description-based filtering can be bypassed through malicious tool names invoked before inspection. This finding motivates a call-timing gate that delays tool invocation until all agent-visible tool metadata has been validated.

In a deliberative poll, once submissions outnumber what anyone will read, some mechanism chooses which arguments each voter sees, acquiring much of the decision; practice delegates it to opaque learned rankers, so a voter cannot recompute or contest the exposure that shaped their vote. We ask whether it can be a published rule over publicly recomputable evidence with parameters held by the voter, treating legibility as an admissibility condition on usable mechanisms, not an objective traded against accuracy. We formalise a poll over bipolar justification sets, judging a slate by reason coverage, the order it arrives in, and captured endorsement mass; we give seven checkable criteria for a civic recommender and a rule meeting them: a one-hop reversed endorsement flow parameterised by a relation-weight function. An agentic simulator records every slate at every vote, over about 17,000 seed-paired runs. Served slates fall 0.035 short of a label-reading ceiling upper-bounding every selection procedure, opaque ones included: any unconstrained ranker's advantage is bounded and small. On coverage alone, with non-degenerate authoring, the rule is indistinguishable from a random slate, a null due to an order-blind, charity-blind instrument; on the other two it leads at every prefix by a margin widening with adversarial pressure and dominates on mass by a factor of 3.3. Once a realistic fraction of submissions carries no reasons, the coverage margin returns and grows. Label-homogeneous flooding collapses completeness from 0.81 to 0.34 under a flat weight policy, only to 0.44 under author-count normalisation, making the weight function a security control worth 10% of completeness. The choice between ranking arms is a position on a coverage-versus-mass frontier, not a fact, the kind of choice only a legible rule can hand to the person it affects. It maps onto an open-source peer-to-peer platform.

Retrieval-augmented generation (RAG) improves the factuality of large language models by grounding responses in external documents, but it also exposes a critical security vulnerability: adversarial documents injected into the knowledge database can enter the context window and steer the model toward targeted incorrect answers. Existing post-retrieval defenses rely on instruction following, parametric knowledge, or text-level consistency, all of which can be imitated or optimized against by adaptive attackers. We propose RAGSentinel, a training-free, label-free defense for black-box RAG systems. RAGSentinel uses a surrogate encoder to measure query-conditioned hidden-state shifts induced by retrieved documents, removes shared topic directions, and filters poisoned documents as geometric outliers from a robust majority consensus. We prove that, under an honest-majority assumption and a representation-level separation condition, RAGSentinel exactly recovers a poison-free majority-sized context. Experiments across three question-answering datasets, three LLM families, and multiple poisoning attacks show that RAGSentinel consistently achieves low attack success rates while preserving competitive accuracy and remaining effective against adaptive attacks with full pipeline knowledge.