Searcharxiv⌕ Search

subject

cs.CR

cs.CR: explore 594 source-linked works published from 2022 to 2026, with original documents and citations.

This collection is a preview while coverage and quality are evaluated.

Search within this collection

Coverage and selection

Includes records with this source-supplied label or an explicit phrase match in their metadata. Matches indicate a mention, not proof that a paper uses a method or tests a material. Source versions are consolidated by DOI.

Sources: arxiv. Collection updated 2026-09-16. Counts describe this index, not the complete source archives.

arXiv2026

Bait-and-Recover: Poisoning Internal Refusal Signals to Defend LLMs against White-Box Editing Jailbreaks

Open-weight large language models face a low-cost white-box threat from representation engineering attacks. Attackers can estimate refusal directions and search for projection-matrix edits that suppress safety alignment while preserving general capabilities, within minutes on a single GPU and without gradient-based training. We propose Bait-and-Recover, a weight-level defense that places a bait adapter where attackers read activations and a paired recovery adapter at the subsequent layer. Trained via gradient routing, this decouples the observation path from the behavior path. By actively poisoning the residual signal used for measurement, Bait-and-Recover disrupts the attacker's edit search, while the recovery layer restores clean downstream computation. Across four open-weight models, our defense raises the minimum refusal rate against white-box edit searches from 16.25% to 71.75% under a strict behavior-preservation budget (KL <= 0.10), with negligible impact on general benchmarks. By invalidating the core measurement assumption of these attacks, observation-path poisoning offers a practical complement to behavior-level safety training.

cs.CR↗
arXiv2026

SAFEGuard: Detect Optimization-Based Jailbreak Attacks Through Harmful Semantic Analysis and Fluency Measurement

Despite the significant efforts devoted to aligning large language models (LLMs) with human values and ensuring safe deployment, recent work has revealed that LLMs remain vulnerable to adversarial jailbreak attacks that can bypass safety guardrails and elicit harmful responses. Many defense methods are proposed to detect jailbreaks but they are limited in their effectiveness to counter wide-range optimization-based jailbreak mechanisms that can yield highly fluency-optimized or harmful semantic obfuscated prompts. To tackle this challenge, we propose a unified detection framework SAFEGuard which incorporates a hybrid fluency measurement based on cross-layer distribution distance and perplexity, and the analysis of harmful semantics through gradient matching. Our method is grounded in a paramount observation: high fluency prompts maintain their malicious intention close to harmful prompts while harmful semantic obfuscated prompts often inject gibberish token sequences. Our evaluation demonstrates that SAFEGuard consistently outperforms state-of-the-art baselines and achieves significant improvement in accuracy across different optimization-based jailbreaks. This underscores the effectiveness of SAFEGuard against evolving jailbreak attacks.

cs.LG↗
arXiv2026

Multimodal Resource-Exhaustion Attacks on Vision-Language Models via Joint Pixel-Prompt Optimization

Resource-exhaustion attacks against autoregressive vision-language models (VLMs) typically assume unimodal threat models, treating the image branch as the primary optimization surface while holding user-visible prompts fixed. Even recent loop-centric variants remain confined to this single-channel paradigm, leaving the exploitation of availability unexplored as a cross-modal optimization problem over jointly controllable input surfaces. We introduce Joint Pixel-Prompt Optimization (JPPO), the first compound adversarial framework elevating the visible prompt to a first-class adversarial variable alongside image perturbations. Under a restricted joint-input threat model, JPPO performs coupled, stagewise optimization over both the pixel and prompt surfaces. This produces synergistic cost amplification, mechanistically distinct from loop-dependent failures, exhibiting negligible loop incidence in our experiments. Evaluating five open-source VLM families on MS COCO and ImageNet under an 8/255 infinity-norm budget, JPPO achieves over 4.6x latency and 5.3x energy amplification on Qwen2.5-VL-7B, and over 36.6x latency with 32.7x energy amplification on BLIP-2. This represents the strongest cost amplification among directly compared baselines while requiring substantially fewer optimization iterations. Ablations confirm this amplification arises from multimodal coordination rather than prompt length or isolated modalities. These findings reveal structural blind spots in current VLM serving defenses, motivating cost-aware robustness evaluation as a first-class security requirement for multimodal deployments.

cs.AI↗
arXiv2026

From Review to Authorization: Key-Isolated Threshold Signing for LLM Agents

Autonomous LLM agents can turn untrusted content into effectful actions such as payments and permission changes. If the same process interprets this content and controls a reusable signing credential, prompt injection can cross the judgment boundary and reach execution authority. We present KITA, a review-to-authorization architecture that keeps the user's personal secret signing key and every threshold signing-key share outside all LLM processes. Under threshold signature unforgeability and our system assumptions, compromising the proposer and fewer than t reviewer-signer domains cannot produce a valid authorization for a new action without signing contributions from t distinct domains. Thus, any such authorization includes a share from an uncompromised domain, bound to the canonical action and released only after authenticated reviewer approval. This establishes execution-bound authorization integrity. We implement the complete reviewer-to-executor path with a structured-output LLM adapter and threshold BLS. Six system tests validate quorum gating and message binding at this interface, while cryptographic microbenchmarks measure the online signing path and its scaling behavior.

cs.CR↗
arXiv2026

EvoSafeHarness: Evolving Model- and Domain-Specific Harnesses for Securing Agents

Large language model (LLM) agents can translate natural-language requests into real-world actions, creating safety risks from both indirect prompt injections and directly harmful requests. Existing system-level safety harnesses are typically designed by experts and applied across heterogeneous models and domains, despite substantial differences in model behavior, utility tolerance, and domain-specific safety requirements. We present EvoSafeHarness, a framework for automatically synthesizing deployment-specific safety harnesses for frozen LLM agents. Given a target model and domain specification, EvoSafeHarness jointly evolves a natural-language policy and executable enforcement logic using behavioral feedback from the target model. Each candidate is additionally evaluated through fresh-context adversarial review to filter rules that depend on benchmark artifacts. Across four agent benchmark families, EvoSafeHarness achieves a stronger safety-utility frontier than fixed expert-designed defenses. On DecodingTrust-Agent, evaluated across fifteen model-domain deployments, it reduces attack success rate from 50.9% to 12.6% for direct attacks and from 40.4% to 7.4% for indirect attacks, with a 3.3-point utility cost. On AgentDojo, it achieves 82.8% utility with 0.0% attack success rate, twice the utility of CaMeL at the same zero-attack-success operating point, while the same harness transfers unchanged to unseen AgentDyn suites. It also achieves the best score on Agent-SafetyBench for every evaluated victim model and maintains an attack success rate below 18% against an adaptive PAIR-style attacker with sixteen attempts per task. Analysis of the synthesized harnesses shows that domain semantics determine which safety relations and trajectory state must be tracked, while model and runtime behavior determine how and where those relations should be enforced.

cs.CR↗
arXiv2026

Structurally Close, Temporally Distant: Measuring Security Exposure in Long-Horizon LLM Agents

Long-horizon LLM agents interact with untrusted content, persistent memory, external state, and sensitive tools. Existing analyses often characterize attacks by the number of execution steps between malicious input and a downstream action. We show that temporal remoteness can overstate security separation in stateful agents. We introduce a provenance-aware execution graph linking agent events through deterministic state, identifier, and tool provenance, and define \emph{influence distance} $\DI$ as the shortest structural path from an untrusted source to a sensitive action. We compare it with \emph{sequence distance} $\DT$, the shortest injection--sink path in the ordered trajectory. Since the influence graph contains every sequence edge, $\DI \leq \DT$; $\Gap=\DT-\DI$ measures the separation hidden by step count. Across 454 injection--sink pairs from 360 long-horizon AgentDojo trajectories over OpenAI's \texttt{gpt-4o-mini} and \texttt{gpt-4o} and Claude's Haiku 4.5 and Sonnet 4.6, $\Gap>0$ for 96.9% of pairs, with a median gap of 9 hops; 91.0% remain decoupled after removing the largest provenance-only edge class. On AgentDojo's banking suite, 33.8% of 231 pairs from 377 trajectories decouple through different provenance mechanisms. Among 274 OpenAI pairs, $\Gap$ does not independently predict attack success after controlling for $\DT$, attack family, and backend ($β_{\Gap}=0.066$, $p=.088$). At matched thresholds $k=2,3$, a deterministic $\DI$-based pre-execution gate blocks five attack sinks missed by a sequence-only gate with no additional benign blocking, although the paired gain is not significant ($p=.0625$). Execution structure therefore reveals proximity hidden by step count and can support targeted runtime intervention. We measure candidate influence pathways rather than causal attribution.

cs.CR↗
arXiv2026

Versioned Transitive Dependency-Closure Binding and Operation-Time Effect Governance for Agent Skills: ClosureBound

Agent Skills combine instructions with files, packages, tools, models, and services, so operational identity can exceed a signed directory. Recursive or lazy dependencies may change while root-level evidence remains valid, and different surfaces may reach the same durable effect. We present ClosureBound, a reference monitor that prevents authorization transfer across material changes to this heterogeneous closure. Its resolver commits typed graph nodes and topology. Each grant binds an exact closure root, effect ceiling, purpose/provenance, validity, and epochs. At durability, it re-resolves closure and state, normalizes the operation into an external-effect IR, and admits it only if a joint witness satisfies every bound. Supported equivalent paths share one ceiling. Assuming complete mediation and discovery, authenticated freshness, sound normalization, cryptographic binding, and authoritative linearization, we establish metadata non-authority, closure determinism, version non-inheritance, effect non-amplification, bound-value freshness, and path invariance. We do not establish program equivalence or remote-service honesty. A provider-free implementation matches 40 frozen lifecycle fixtures; 18 kernel contracts and six mutants cover binding and downgrade cases. Full-profile exploration reaches 84,608 states and 530,752 transitions without a declared invariant violation; six weakened profiles yield witnesses. A lexical audit of 549 public Skills (4,872 unique files) finds that 21 of 526 roots with bundled files name every non-manifest path verbatim, 67 contain links resolving outside their roots, and no root declares a frontmatter dependencies field. These observations motivate conservative closure discovery and define concrete targets for broader runtime, interoperability, efficacy, and production validation.

cs.CR↗
arXiv2026

Robustness Evaluation and Detection of Transferable Adversarial Attacks in ML-Based NIDS

Machine learning-based network intrusion detection systems (ML-based NIDS) are vulnerable to adversarial evasion, where malicious samples are perturbed to evade detection and be misclassified as benign. Despite growing research on adversarial attacks and defenses for ML-based NIDS, comparative evaluations of multiple attack types, detection models, and defense strategies under a common setting remain limited. In this paper, we evaluate eight adversarial evasion attacks, fifteen detection models, and three representative defense strategies using the NF-UQ-NIDS dataset, which includes recent traditional and IoT network traffic with twenty distinct attack categories. The evaluation compares model performance on clean test data and on robustness evaluation sets that include adversarial samples, analyzes attack success consistency across models, and examines the effect of defense strategies on adversarial robustness. To support model comparison, we introduce the Robustness Index (RI), a compact comparative metric that rewards high balanced accuracy and macro-F1 score computed on the robustness evaluation set while penalizing high attack success rate (ASR). We further present AR-NIDS, a two-stage framework that uses an adversarially trained ensemble to distinguish normal, attack, and adversarial samples, followed by an adversarial attack classifier to identify the attack type. Under the evaluated transfer-based setting, the proposed adversarially trained ensemble achieves the strongest overall trade-off between classification performance on the robustness evaluation set and evasion resistance, reducing the average ASR from 0.41 to 0.03 and achieving an RI of 0.98.

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

SemVul: Semantic-Enhanced Graph Neural Networks for Code Property Graph-based Vulnerability Detection

Vulnerabilities in source code are often the root cause of cyberattacks worldwide, as attackers exploit weaknesses in software to gain unauthorized access, steal data, or disrupt services. In this study, we evaluated existing research approaches and propose SemVul, a vulnerability detection pipeline that demonstrates better generalization and higher accuracy in learning vulnerable code patterns. We propose a Code Property Graph-based vulnerability-detection approach combined with semantic-level enhancement, enabling the model to capture both the program's structural flow and the semantic meaning of the code. Our approach integrates both node-level and edge-level semantic embeddings using pre-trained code embedding techniques. We systematically evaluate multiple GNN architectures on publicly available benchmark datasets. SemVul is generic with respect to the programming language and supports multiple architectures. By integrating structural and semantic information, the proposed approach improves vulnerability detection performance. Our results show that SemVul outperforms existing approaches and provides better generalization.

cs.SE↗
arXiv2026

Beyond the Prank: The Hidden Expertise of TSS Scambaiters

This research studies how Technical Support Scams (TSS) are being countered by a uniquely dedicated community of volunteer counter-fraud operatives. Using a careful subject selection strategy, we interviewed 17 individuals who actively engage in TSS scambaiting activities in order to obtain insight into their motivations, the operational methods of the scammers they combat, the undocumented nuances of effective scambaiting action, and the various challenges scambaiters face. In our analysis, we find a community rich not only with insight into offenders, but with technical and operational expertise that is often lacking in research efforts targeting these same populations. At the same time, we find key areas where the community could be better supported and enabled. We discuss the implications of our findings for both future research and community protection strategies.

cs.CR↗
arXiv2026

Counter-Swarm Doctrine: Containing Coordinated Agent Intrusions

Agents can turn shared infrastructure into a channel for coordinated intrusion. The Hugging Face incident and a separate public-wiki investigation show why a security assessment may need evidence from several executions and the artifacts they leave behind. We argue that the operational unit of defence should be a revisable coordination episode linking observed transfers, task authority, and response history. The central research problem is prospective episode discovery: finding which actions belong together before an evaluator supplies their membership. We define unsanctioned coordination relative to collaboration and delegated-authority policy, connect storage-mediated coordination to stigmergy, and specify the evidence needed to distinguish influence from common causes. First-contact signals are one possible input to discovery; the design also follows inherited state and later use. A proposed evaluation compares isolated actions, rolling windows, known groups, and prospectively discovered episodes at matched review cost and false-alert workload. It measures harmful outcomes across all assigned population runs and tests recurrence after channel closure and state quarantine. A checksum-verified reconstruction of the public wiki export separates the decline in retained writes from later administrative cleanup. The contribution is an incident-grounded position, descriptive analysis, and evaluation design. It makes the recommendation to monitor across executions testable without claiming a new detector or a measured containment benefit.

cs.CR↗
arXiv2026

SCRIPTIOC-BENCH: A Benchmark for Recognizing Actionable Threat Intelligence from Script-Based Malware using LLMs

Script-based malware remains a prevalent attack technique. These scripts often contain indicators of compromise (IOCs) that provide actionable threat intelligence. However, statically recovering such indicators is challenging, as relevant values may be dispersed or transformed within code. Although large language models (LLMs) have shown promise in security analysis, their ability to recover IOCs from malicious scripts remains underexplored. We present SCRIPTIOC-BENCH, a benchmark for measuring static IOC extraction capability on real-world malicious scripts. The benchmark comprises 634 manually verified JavaScript, PowerShell, and VBScript malware samples covering four IOC types (URLs, domains, IP addresses, and filesystem artifacts). We further stratify ground-truth IOCs by recovery level, distinguishing directly exposed indicators from those requiring decoding or reconstruction. Using this benchmark, we evaluate a broad range of proprietary and open-weight LLMs and show that IOC recovery without execution remains challenging across model scales: the strongest model reaches only 65.4 F1. To characterize how recovery fails, we introduce a false-positive taxonomy and use it to compare the error profiles of the evaluated models. We further study two mitigations on a small open-weight model, deterministic string utilities and task-specific adaptation, finding that they provide complementary recovery gains, raise precision, and shift errors toward sample-grounded mismatches.

cs.CR↗
arXiv2026

Scratchy: Visual-Scratchpad Multimodal Reasoning for Cryptographic Proof Generation in EasyCrypt

Large language models (LLMs) have recently made substantial progress in formal proof generation, yet presenting distinctive challenges in cryptographic area. Computational security arguments posit that a valid proof must coordinate probability, adversarial games, invariants, assumptions and bounds, which can be provided by a machine-checked framework named EasyCrypt. Although all objects may appear in available context, LLMs still struggle because proof-theoretic dependencies are typically implicit in a linear representation and distributed across multiple programs. So, this paper presents Scratchy, a visual-scratchpad approach that exposes these dependencies for multimodal generation. Given the natural-language security description, with formal context and target propositions, the proof objects can be normalized into a typed proof-relation graph. Then a structure-preserving visual compiler transforms the graph into the formula-rich visual proof state that guides a multimodal model in generating the EasyCrypt proof. Also, the Scratchy-eval, a 114-task dataset derived from reliable official EasyCrypt files, has been introduced. It contains 64 security-form proof generations and 50 multiple-choice knowledge tests. After a series of evaluations, covering semantic grounding, relational invariants, and game reductions, classical LLMs like GPT-5.6-Sol and Claude-Opus-5 have gained a clear advantage from Scratchy's structured visual proof states. This contrast suggests that explicit proof structure can make the improvement and multimodal proof-state representation as a promising direction for computer-aided cryptography.

cs.CR↗
arXiv2026

It is Not Yet Another Tool: Creating and Deploying an Agentic AI Companion in a Security Operations Center

Security Operations Centers (SOCs) process large amounts of tickets, most of which are low-interest events not worthy of further investigation. The repetitive nature of this task and similarity of the vast amounts of tickets make it a prime candidate for generative AI-based automation. We created and deployed an agentic AI companion utilizing large language models through fieldwork within a SOC for over one year. The design of the SOC AI companion was driven by researchers' participation and interactions within the SOC's daily work. SOC analysts were invited to use it during the last four months of the fieldwork. We analyzed the analysts' usage of the companion and found that in more than 90% of the cases the companion's outputs were reused by analysts in the ticket's closing report. Our results showed that when designed "in the trenches" with the intended users, a SOC AI companion can go beyond being yet another tool, but rather a system that co-evolves with its human users as it traverses through the various types of workloads. Analysts naturally started to shape the AI companion's behaviors to fit their particular needs. Our data show that the more human analysts shape the AI companion's behaviors, the more they become comfortable trusting the output from the AI system, resulting in improved productivity.

cs.CR↗
arXiv2026

MARS: Detecting Unauthorized Variable Manipulations in Multi-Application PLC Runtimes

Programmable Logic Controllers (PLCs) increasingly run multiple applications alongside the main control program, with shared access to PLC variables. Yet, Industrial Control System (ICS) defenses primarily detect malicious updates by checking whether variable values violate expected bounds, without considering which application performed the update. A malicious application can exploit this gap by modifying variables within normal bounds while still driving the physical process toward an unsafe state. Even when such manipulation is detected, operators cannot identify the responsible application because PLCs do not associate variable updates with application identity. We present MARS, an automated framework for application-level authorization and attribution of PLC variable manipulations. MARS profiles applications on an isolated virtual PLC (vPLC) to derive application-specific variable-access policies and uses a shadow vPLC during operation to attribute production-PLC updates to individual applications without instrumenting the production controller. MARS also detects manipulations that occur only on the production PLC and therefore have no corresponding update on the shadow vPLC. We evaluate MARS on manufacturing, chemical, and water-treatment systems against attacks in which unauthorized applications manipulate PLC variables while remaining within normal bounds. Our results show that MARS detects these manipulations and identifies the responsible application.

cs.CR↗
arXiv2026

Statistical Inference for Privatized Data with Unknown Sample Size

We develop both theory and algorithms to analyze privatized data in unbounded differential privacy (DP), where even the sample size is considered a sensitive quantity that requires privacy protection. We show that the distance between the sampling distributions under unbounded DP and bounded DP goes to zero as the sample size $n$ goes to infinity, provided that the noise used to privatize $n$ is at an appropriate rate; we also establish that Approximate Bayesian Computation (ABC)-type posterior distributions converge under similar assumptions. We further give asymptotic results in regimes where the privacy budgets vary, establishing similarity of sampling distributions as well as showing that the MLE in the unbounded setting converges to the bounded-DP MLE. To facilitate valid, finite-sample Bayesian inference on privatized data under unbounded DP, we propose a reversible jump MCMC algorithm which extends the data augmentation MCMC of Ju et al. (2022). We also propose a Monte Carlo EM algorithm to compute the MLE from privatized data in both bounded and unbounded DP. We apply our methodology to analyze a linear regression model as well as a 2019 American Time Use Survey Microdata File which we model using a Dirichlet distribution.

math.ST↗
arXiv2026

Proof verification by polynomial Fingerprinting

To cater to the needs of fast verification for mathematical proofs, we describe a method to encode formal sentences in $2 \times 2$ - matrices over multivariate polynomials with integer coefficients. This correspondence is homomorphic: usual proof-steps like modus-ponens or variable substitution in terms and formulae become operations with matrices. By evaluating the polynomial variables in random elements of a suitably chosen finite field, the proof is replaced by a numeric sequence. Only the values corresponding to axioms and tautologies have to be computed from scratch. The values corresponding to derived formulas are computed from the values corresponding to their ancestors by applying the homomorphic properties. The polynomial matrix corresponding to the conclusion of the proof is also evaluated in the chosen random values. If the last term of the numeric sequence equals the evaluation of the conclusion, by the Schwartz-Zippel Lemma, the proof is with high probability correct.

math.LO↗
Compare source metadata on this page
WorkPublishedSource identifierSource
Bait-and-Recover: Poisoning Internal Refusal Signals to Defend LLMs against White-Box Editing Jailbreaks2026-09-052609.05794arxiv
SAFEGuard: Detect Optimization-Based Jailbreak Attacks Through Harmful Semantic Analysis and Fluency Measurement2026-09-052609.05850arxiv
Multimodal Resource-Exhaustion Attacks on Vision-Language Models via Joint Pixel-Prompt Optimization2026-09-052609.05889arxiv
From Review to Authorization: Key-Isolated Threshold Signing for LLM Agents2026-09-052609.05901arxiv
EvoSafeHarness: Evolving Model- and Domain-Specific Harnesses for Securing Agents2026-10-142609.05903arxiv
Structurally Close, Temporally Distant: Measuring Security Exposure in Long-Horizon LLM Agents2026-09-052609.05911arxiv
Versioned Transitive Dependency-Closure Binding and Operation-Time Effect Governance for Agent Skills: ClosureBound2026-09-052609.05920arxiv
Robustness Evaluation and Detection of Transferable Adversarial Attacks in ML-Based NIDS2026-09-052609.06012arxiv
Evaluating Deep-Search Agents under Hierarchical Web Evidence Poisoning2026-09-172609.06027arxiv
SemVul: Semantic-Enhanced Graph Neural Networks for Code Property Graph-based Vulnerability Detection2026-09-052609.06035arxiv
Beyond the Prank: The Hidden Expertise of TSS Scambaiters2026-09-052609.06103arxiv
Counter-Swarm Doctrine: Containing Coordinated Agent Intrusions2026-09-052609.06140arxiv
SCRIPTIOC-BENCH: A Benchmark for Recognizing Actionable Threat Intelligence from Script-Based Malware using LLMs2026-09-052609.06149arxiv
Scratchy: Visual-Scratchpad Multimodal Reasoning for Cryptographic Proof Generation in EasyCrypt2026-09-052609.06226arxiv
It is Not Yet Another Tool: Creating and Deploying an Agentic AI Companion in a Security Operations Center2026-09-052609.06250arxiv
MARS: Detecting Unauthorized Variable Manipulations in Multi-Application PLC Runtimes2026-09-052609.06306arxiv
Statistical Inference for Privatized Data with Unknown Sample Size2026-09-042406.06231arxiv
Proof verification by polynomial Fingerprinting2026-09-042506.21114arxiv

These are bibliographic comparisons, not experimental rankings. Follow the original document for methods and conditions.