Search arXiv⌕ Search

SEARCH · Search arXiv

Search Search arXiv

Search indexed arXiv papers on artificial intelligence, large language models, computer vision and robotics. Read source abstracts and follow links to arXiv.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 55 records · Page 3Linked to original sources
arXiv2026

Safety Does Not Compose: Non-Decaying Loop State for Autonomous LLM Agents

Large language model agents are increasingly deployed as autonomous loops. Starting from one human goal, such a system repeatedly discovers work, plans, executes tool calls, verifies outcomes and persists state across many unattended iterations. The agent safeguards in wide use, however, are defined over a single trajectory, and their safety state is re-initialized when the next trajectory begins. We show that this is a failure of composition rather than an implementation detail. Our central result is a separation: against an attack whose evidence is fragmented across several iterations, every trajectory-scoped monitor has a true-positive rate equal to its false-positive rate, however expressive it is, because the evidence it would need never appears in the window it sees, whereas a monitor retaining cross-iteration state separates the two perfectly. We further show that the obvious repair of carrying a geometrically decaying risk score is insufficient, because the cooling-off period a patient adversary must wait is a constant that does not grow with the horizon $N$. We then present LoopHarness, which restores a persistent, non-decaying safety state at the loop level. Under mediated commits and an arbiter detection floor $δ_M$, it bounds the expected number of unauthorized irreversible actions by $B+m-1+m/δ_M$, a constant in $N$, of which the $B+m-1$ term is decided by a model-free rule and therefore survives a fully colluding verifier. We give a complete evaluation protocol on native Agent-SafetyBench tasks with paired clean and attacked episodes, an outer-state attack suite whose decisive evidence exists only across iterations, per-module ablations, and an adaptive white-box red team.

cs.CR↗
arXiv2026

Bayesian and Multi-Objective Decision Support for Incident Mitigation in Cyber-Physical Systems

Cyber-physical systems increasingly rely on interconnected physical and digital systems whose security incidents can escalate rapidly into safety and operational failures. Existing decision-support approaches struggle to support incident response because they rely on static assumptions, incomplete vulnerability data, and single-objective risk models that do not adequately capture trade-offs between attack success likelihood, impact severity, and system availability. This paper proposes an adaptive decision-support framework for incident mitigation in cyber-physical systems that integrates hierarchical Bayesian Network modelling, confidence-calibrated exposure estimation, and multi-objective optimisation into a unified, adaptive pipeline. The framework constructs probabilistic models from system architecture and vulnerability data, incorporating complementary vulnerability scores under epistemic uncertainty as conservative, uncertainty-aware reporting metrics for supporting downstream risk assessment. Mitigation strategies are explored as countermeasure portfolios and refined using multi-objective optimisation to identify Pareto-optimal trade-offs suitable for incident response scenarios. Frequency-based heuristics are applied to prioritise mitigation actions across optimisation runs. The framework is evaluated on three representative cyber-physical attack scenarios, demonstrating its ability to adapt to evolving threats and provide actionable decision support under operational constraints, with the aim of enhancing the resilience of cyber-physical systems.

cs.CR↗
arXiv2026

Impact Analysis of Speech Representation Learning Models for Acoustic Side-Channel Attack

Acoustic side-channel attacks (ASCA) on keyboards have gained increasing attention, yet impact of speech representation learning models in ASCA remains unexplored. Addressing this, we introduce KEYAC, a dataset designed to analyze representation generalization for ASCA under both standard and VoIP codec settings. On KEYAC, we evaluate six representation learning models under zero-shot and partial fine-tuning settings using fully connected and convolutional networks. Results show that while partial fine-tuning improves performance, models struggle to generalize across VoIP codecs. We hypothesize this limitation stems from inadequate modeling of nonlinear feature interactions in conventional fine-tuning architectures. To address this, we employ Kolmogorov-Arnold Networks (KAN) for fine-tuning. Empirical results show that KAN-based fine-tuning consistently outperforms the baselines and establishes a new state-of-the-art on KEYAC.

cs.CR↗
arXiv2026

Compared to What? A Human-Anchored Security Benchmark for LLM-Generated Infrastructure-as-Code

Large language models increasingly author Infrastructure-as-Code (IaC), where one insecure default is provisioned straight into production. Prior evaluations report vulnerability counts for models only, and so cannot say whether models are worse than the engineers they assist. We present GenIaC-SecBench: 100 deployment scenarios across 12 model configurations from six vendors, open and closed weights, yielding 1,196 artifacts scanned by three policy engines (Checkov, Trivy, KICS) at complete coverage. Crucially we scan 634 human-authored IaC templates with the identical toolchain, giving the first size-matched human security baseline for this task. Vulnerability density is strongly inverse to artifact size (Spearman $ρ=-0.55$, $p<10^{-77}$), so unmatched comparisons measure size, not security. Size-matched, every configuration exceeds the human baseline at $3.21\times$ to $3.87\times$, and the gap widens as tasks get simpler ($4.9\times$ at one resource, $1.4\times$ at twenty or more). A majority of scenarios prescribe a security state rather than specifying function alone, so we stratify by prompt class: pooled the gap is $3.50\times$, and excluding every scenario that explicitly requests an insecure configuration still leaves all configurations above baseline ($2.4\times$ to $4.2\times$). The corpus cannot isolate unprompted default posture, and we say so. Decomposing "reasoning" into standard generation, prompted chain-of-thought, and vendor extended-thinking APIs, extended thinking beats prompted CoT ($-12.0\%$, $p=0.0013$) while prompted CoT alone is indistinguishable from standard ($-1.3\%$, n.s.); it consumes under $1\%$ of the output budget, bounding the effect. Two negative results: more deployable models are not more vulnerable ($r=0.158$, $p=0.625$), and complete-case Friedman is uncomputable here, motivating Skillings-Mack. All code and data are released.

cs.CR↗
arXiv2026

SingProbe Technical Report

We present SingProbe, an open intrinsic guardrail framework for generation-time monitoring of LLMs. Intrinsic guardrails reuse hidden states already produced by the base model during autoregressive decoding, rather than relying on an independent model to repeatedly process generated text. While this route has been explored in industrial systems, the community lacks a broadly reusable open stack that combines cross-model guard adaptations, unified training methods, serving integrations, and systematic evaluation resources. SingProbe is designed to provide this missing layer and uses a lightweight probe to continuously produce query-intent, response-safety, and hallucination-risk signals during decoding. This report describes the full intrinsic-guardrail stack: training methods, serving integrations with SGLang and vLLM, and adapted guard models for 29 open-source base models across diverse families and scales. We also introduce SingStreamBench, a benchmark that measures whether streaming guardrails remain inactive on benign prefixes while promptly detecting emerging unsafe content. Across evaluations of safety, streaming detection, hallucination detection, false-positive robustness, online monitoring, and runtime overhead, SingProbe provides performance competitive with, and in several settings stronger than, state-of-the-art standalone guardrails and specialized hallucination detectors, while adding less than 0.5% serving overhead in our implementation. Beyond passive monitoring, we show that intrinsic guard signals can guide constrained decoding and selectively activate medical-risk interventions in SingProbe-Med. By open-sourcing our infrastructure, training methods, and model adaptations, we aim to facilitate the broader adoption and deployment of intrinsic guardrails, as well as further research in this direction.

cs.CR↗
arXiv2026

Staying on the Attack Path: Structured State for Long-Horizon Automated Penetration Testing

Large language model (LLM) based agents are increasingly applied to cybersecurity tasks such as vulnerability discovery and automated penetration testing. On long-horizon security tasks, however, such agents remain limited by context forgetting and intent drift: early critical facts and causal reasoning chains are lost over extended interactions, and the agent falls into aimless, repetitive exploration. This paper proposes Intentest, an intent-graph-guided automated penetration testing agent that externalizes long-horizon state from the LLM's context window onto a persistent fact-intent directed acyclic graph (DAG), thereby substantially reducing invalid transitions. We evaluate Intentest on automated penetration testing of web applications, a representative long-tail task in cybersecurity. In the DAG, verified network states are stored as immutable fact nodes, and exploration directions are constrained as intent edges bounded by predecessor facts. The system adopts a three-layer architecture, in which the fact-intent mapping layer maintains the global state, the task scheduling and allocation layer ensures execution stability through two-phase degradation recovery and multi-dimensional adaptive load balancing, and the intent retrieval and prediction layer provides tactical priors through a top-down five-stage filtering algorithm. On a benchmark of real CTF challenges covering more than ten vulnerability types across three difficulty levels, Intentest achieves an overall success rate of 88.2% and a success rate of 75.0% on hard tasks, improving over the baseline by approximately 44 and 50 percentage points. Ablation experiments further show that the intent retrieval and prediction reduce the average number of rounds on successful medium and hard tasks by about 33% and 48%, respectively, without changing the set of solvable tasks.

cs.CR↗
arXiv2026

FlyBlind: Cross-Slice Timeliness Attacks on UAV Situational Awareness over 5G

Beyond Visual Line of Sight (BVLOS) Uncrewed Aerial Systems (UAS) operating over 5G Standalone (SA) networks use a shared User Plane for both command-and-control (C2) data and video feedback. Operators assess link quality through latency and availability, relying on soft isolation between network slices. However, the risk that an authorized co-tenant could make the Ground Control Station (GCS) state outdated without disrupting the connection remains underexplored. This work introduces FlyBlind, a timeliness attack in which an authorized co-tenant on a neighboring slice maintains legitimate uplink demand, causing state aging at the GCS without a rogue gNB or direct interference with C2 traffic. Our key insight is that, under soft isolation, sharing idle resources turns authorized competition for grants into state aging that conventional link monitors fail to detect. We formalize this effect, termed Silent State Staleness, as a falsifiable false-healthy predicate. On a dedicated testbed, telemetry age at the GCS saturates at approximately 12 seconds, with discrepancies of tens of meters between the GCS position estimate and ground truth, while the one-way delay (OWD) p99 remains in the tens of milliseconds, availability exceeds 99.9%, and the vehicle keeps operating in GUIDED mode without triggering failsafe mechanisms. These findings indicate that, in deployments with asymmetric uplink enforcement, verifying state freshness at the destination is essential rather than relying solely on link health.

cs.CR↗
arXiv2026

Position Matters: Feature Inversion Attacks in ViT Split Inference with Token Reduction and Shuffling

Vision Transformers (ViTs) are increasingly used in split-inference systems, where edge devices transmit intermediate token representations to a remote cloud. In this setting, token reduction lowers computation and communication costs, while token shuffling disrupts the spatial organization of the transmitted tokens, potentially limiting information leakage. However, their privacy benefits remain unclear against feature inversion attacks, which attempt to reconstruct the input from the transmitted embeddings. In this work, we show that, despite disrupting the spatial structure required by conventional reconstruction attacks, transmitted token embeddings retain substantial positional information. Based on this observation, we introduce the Spatially Aligned Reconstruction Attack (SARA), a unified pipeline that predicts token positions, restores their spatial layout, reconstructs missing embeddings using a feature-space masked autoencoder, and recovers the input image. Our results demonstrate that token shuffling provides only apparent privacy, as SARA largely reconstructs the original token organization. Token reduction offers stronger protection, but significant leakage persists when the retained tokens preserve sufficient semantic and positional information. Finally, we introduce a lightweight edge-side defense that removes positional embeddings and progressively adapts the edge-side transformer blocks through knowledge distillation. It substantially reduces attack performance against SARA, while preserving downstream task accuracy and requiring no changes to the cloud-side model.

cs.CR↗
arXiv2026

Evaluating Deep-Search Agents under Hierarchical Web Evidence Poisoning

Search-augmented LLM agents are increasingly used for consumer decisions, making them vulnerable to Generative Engine Optimization (GEO) poisoning. Existing benchmarks largely measure whether manipulated content is retrieved or endorsed, but do not track whether an agent verifies suspicious evidence, revises adopted claims, or recovers before producing its final recommendation. We introduce HAE-GEO, a benchmark that tracks the full trajectory from exposure to recovery under progressively more persuasive Web poisoning. Agents interact via a multi-turn Search-Scrape interface across three attack levels (L1 direct assertion, L2 contextual camouflage, and L3 apparent corroboration), supported by a controlled corpus of 72,039 clean pages and 770 poisoned pages per level spanning 8 product categories and 154 brands. Evaluation combines deterministic behavioral measures with six semantic rubric dimensions. Evaluating 10 agents, we find three recurring patterns: evidence recognition degrades under the corroboration trap; agentic search improves final resistance without improving evidence recognition or utility; and defense prompting increases verification, yet rarely converts verification into recovery.

cs.CR↗
arXiv2026

Moirae: A Multimodal Agent Collaborative Framework for Dynamic Android Malware Detection

The Android ecosystem faces persistent and rapidly evolving malware threats. Existing machine learning detectors are vulnerable to concept drift because they rely on implementation-specific features whose distributions change over time. Large language models (LLMs) offer strong semantic understanding and zero-shot reasoning, but current LLM-based detectors typically depend on code-centric or single-dimensional evidence, making them susceptible to obfuscation and limiting comprehensive behavior analysis. We present Moirae, a multimodal agent collaborative framework for dynamic Android malware detection. Moirae dynamically collects multimodal runtime evidence and employs ReAct-based specialized agents to analyze complementary behavioral views. The detection process begins by identifying visual deception cues, modeling UI state transitions, and integrating runtime API behaviors to fuse multi-dimensional evidence across user-visible interfaces and hidden backend operations. Experiments on temporally and distributionally unseen datasets show that Moirae achieves an accuracy of 90.06\% without fine-tuning, outperforming state-of-the-art baselines and demonstrating strong zero-shot generalization against Android malware concept drift.

cs.CR↗
arXiv2026

Curvature Cryptanalysis of Smooth Transformer Feed-Forward Networks

We show that smooth two-layer feed-forward networks (FFNs) expose an additional structural model extraction channel under a chosen-input raw-output oracle at the FFN branch; consider transformer FFN branches with GELU or SiLU activations under chosen-input raw-output access, without access to parameters, gradients, or internal activations; exploit a second-order leakage channel in which projected input Hessians form different mixtures of the same hidden symmetric rank-one factors induced by the FFN input weights. We formalize resulting Hessian collection as a partially symmetric decomposition to establish conditions for local identifiability and stability to exploit vector-output stencil reuse to reduce the structural query cost by a factor of 16. On independently trained CIFAR-10 vision transformers, only 16 projected Hessians, corresponding to 8193 black-box queries, recover the hidden FFN directions with average absolute cosine alignment above 0.91, with 86.7.1 % of GELU and 91.1 % of SiLU directions exceeding 0.90 alignment. Recovery remains high across independently trained models, repeated extraction runs, and all transformer blocks. The recovered structure supports functional extraction too. Keeping the recovered directions fixed and fitting only the remaining FFN parameters yields high-fidelity substitutes with more than 92 % top-1 agreement, while test accuracy remains within 1.24% and 0.57% of the GELU and SiLU targets. Output rounding and Gaussian noise substantially reduce recovery under a fixed attack configuration, but adapting the finite-difference step restores average alignment to 0.9188 and 0.9081. This is an end-to-end path from black-box second-order observations to hidden FFN-structure recovery and functional replacement. Under the stated oracle model, smooth FFN curvature exposes internal parameter geometry that behavioral fidelity alone cannot reveal.

cs.LG↗
arXiv2026

Crossing the Streams: SSH Plaintext Recovery via a Common Compression Context in Multiplexed Channels

SSH is the standard protocol for secure remote administration of servers. At the transport layer, SSH uses the Binary Packet Protocol (BPP) for encrypted and authenticated communication. Above this, the SSH Connection Protocol multiplexes one or more logical channels over a single connection, supporting interactive shells, port forwarding, and related functionality. We show that SSH channel multiplexing creates a previously unrecognized compression side channel: all channels on a connection share the same compression context. When compression is enabled, an attacker can inject partially chosen plaintext into a channel and observe the length of the resulting ciphertext on the network. This enables an adaptive chosen-plaintext attack that recovers secrets from one channel by interacting with another. While related attacks such as CRIME and BREACH have been studied extensively for HTTP over TLS, this is, to our knowledge, the first compression side-channel attack on SSH and the first SSH analysis to consider a combined passive eavesdropper and web attacker threat model. We further demonstrate the attack in three different application scenarios and evaluate its effectiveness under varying levels of protocol noise. We find that, in the lowest-noise scenario, an 8-character secret over a 26-letter alphabet can be recovered using at most 276 guesses. Finally, we analyze the SSH ecosystem for compression support and other implementation characteristics that influence the practical efficacy of the attack.

cs.CR↗
arXiv2026

Repeat-After-Me: Black-Box Adaptive Visual Prompt Injection

Prompt injection is widely recognized as a major security threat to AI agents that interact with untrusted external data, such as websites, documents, and emails. Prior work has shown that, in the text domain, black-box prompt injection can achieve near-perfect attack success rates (ASRs). In the image domain, however, existing visual prompt injection methods are substantially less effective in attacking frontier commercial VLMs for materially harmful behavior. Achieving such outputs is hard because it requires a long and/or format-compliant target string, such as a precise, parseable native tool call with exact function names and arguments. We present Repeat-After-Me, a black-box adaptive visual prompt injection attack that can reveal personally identifiable information or make malicious tool calls. Across both open-weight and commercial frontier VLMs, including Qwen3.6-27B and GPT-5.5, our method achieves ASRs exceeding 82% and 47%, respectively, under a realistic setting in which the benign user prompt is semantically unrelated to the injected task and does not verbally authorize it. Our optimized injection has non-trivial attack transferability across commercial VLMs and benign samples. We show our attack works in cases where adaptive textual prompt injection fails. In a real-world OpenClaw agent connected to Discord, an untrusted user can use a minimally injected image from our attack to overwrite TOOLS.md, enabling future sensitive behaviors like remote code execution and secret exfiltration. We discuss potential defenses.

cs.CR↗
arXiv2026

An Empirical Analysis of CodeQL False Positives and Query Refinements for Java Vulnerabilities

Static application security testing (SAST) tools help developers find vulnerabilities before deployment, but false positives create substantial triage effort. We study whether CodeQL false positives in Java security analysis form recurring, explainable patterns that can be reduced by refining the analysis. We run CodeQL's Java security query suite on 167 CVE instances from 110 projects, focusing on the ten queries with the highest false positive rates. We manually review 500 sampled false positive paths and locations and construct a source-level taxonomy. The five categories are Missed Path Constraint or Sanitization (36.6%), Benign Execution Context (29.4%), Missing Trust Boundary Modeling (27.6%), Imprecise Concurrency Modeling (5%), and Imprecise Sink Modeling (1.4%). Guided by these findings, we implement CodeQL refinements that detect and filter recurring false positive patterns at the query level. The refinements remove 81.8% of reviewed false positives. Across the full selected-query dataset, they remove 15.8% of reported paths and locations while retaining 7 of 8 true positives. This shows that many false positives can be reduced in the analysis, although fixed refinements often depend on project-specific context. To address this generalization gap, we evaluate whether agentic coding tools can adapt refinement patterns to new projects. Given our patterns as templates, the two tools succeed on 56% and 62% of tasks, with query compile-pass rates above 90%. Without this guidance, both succeed on only 28%, while compile rates fall to 30-36%. These results support a refinement-oriented SAST workflow in which recurring false positives are modeled in CodeQL queries and automatically adapted to different project contexts, reducing repeated triage.

cs.SE↗
arXiv2026

Understanding the (In)Security of Vibe-Coded Applications

Recent advances in large language models (LLMs) have enabled vibe coding, an emerging software development paradigm in which users create applications primarily through natural-language interactions with AI agents. Due to its low barrier to entry, vibe coding is rapidly gaining adoption in practice. Unlike conventional AI-assisted programming, where developers remain responsible for implementation and code review, vibe coding delegates a substantial portion of the development process to AI systems. This shift raises a fundamental question: how (in)secure are applications developed through vibe coding? In this paper, we conduct a systematic study of the security of real-world vibe-coded applications. We collect 9,041 open-source applications developed using popular AI agents (Claude Code and Lovable), and audit 200 publicly deployed applications, uncovering 1,186 vulnerabilities. Our study of these applications and vulnerabilities reveals several key findings: (1) insecurity is the norm rather than the exception: 91.0\% of audited applications contain at least one vulnerability, and 65.77\% of identified vulnerabilities are rated Critical or High severity, concentrated in broken access control, injection, and authentication failures; (2) these vulnerabilities are traceable to eight recurring failure modes rooted in three systematic limitations of AI agents: memory defects, objective defects, and knowledge defects; and (3) while improved agent harness and prompting strategies can reduce the incidence of vulnerabilities, they do not eliminate the underlying security risks. Overall, our study provides an empirical understanding of the security landscape of vibe-coded applications and lays the groundwork for addressing the security risks in the growing delegation of software development to AI systems.

cs.CR↗
arXiv2026

Networked Multi-Resource Defense Capabilities in a General Lotto Game

Ensuring the security of complex systems involves the strategic allocation of defensive resources to prevent various types of attacks from succeeding. A defender often has multiple types of defensive assets at its disposal, where it must decide how to optimally deploy their heterogeneous capabilities across different attack types. In this paper, we formulate a multi-resource allocation problem in the form of a General Lotto game where a defender possesses various types of resources. A feature that we introduce is that their individual effectiveness against different types of attacks is characterized by a network weight matrix. In our analysis, we derive upper and lower bounds on the performance of the defender, and provide numerical evidence suggesting that they are tight. For the case of two attack types, we analytically prove that the bounds coincide, establishing an exact equilibrium characterization. We then numerically compare our proposed networked multi-resource architecture to an independent-defense benchmark from the existing literature. These results highlight fundamental and tractable structures underlying multi-attack-type defense problems.

cs.GT↗
arXiv2026

A Systematic Review of Security Communication Strategies: Guidelines and Open Challenges

Cybersecurity incidents such as data breaches have become increasingly common, affecting millions of users and organizations worldwide. The complexity of cybersecurity threats challenges the effectiveness of existing security communication strategies. Through a systematic review of over 3,400 papers, we identify specific user difficulties including information overload, technical jargon comprehension, and balancing security awareness with comfort. Our findings reveal consistent communication paradoxes: users require technical details for credibility yet struggle with jargon and need risk awareness without experiencing anxiety. We propose seven evidence-based guidelines to improve security communication and identify critical research gaps including limited studies with older adults, children, and non-US populations, insufficient longitudinal research, and limited protocol sharing for reproducibility. Our guidelines emphasize user-centric communication adapted to cultural and demographic differences while ensuring security advice remains actionable. This work contributes to more effective security communication practices that enable users to recognize and respond to cybersecurity threats appropriately.

cs.CR↗