Searcharxiv⌕ Search

SEARCH · Searcharxiv

Search Searcharxiv

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 19 recordsLinked to original sources
arXiv2026

Same Request, Different Boundary: Evaluating Cybersecurity Assistance across Conversational Contexts

Cybersecurity assistance is dual use, requiring decisions about when to answer, limit, or refuse a request. Prior work studies refusal evaluation and conversational jailbreaking, but how conversational histories shift assistance boundaries in both directions for identical cybersecurity requests remains underexplored. We introduce 3R-Bench, a benchmark of 150 real-world cybersecurity requests evaluated across eight LLMs under direct prompting and two predefined conversational settings. We separately code refusal, bounded assistance, compliance, and provider blocking to characterize delivered assistance. On 400 matched ambiguous request-model pairs, compliance is 75.3% directly, 57.8% after refused precedent, and 81.0% after accepted precedent. A smaller repeated-generation panel reproduces the refusal-acceptance gap. On a balanced 75-request panel reusing the direct baseline, compliance decreases from 74.2% directly to 57.2% after neutral task-decomposition dialogue and 60.3% after failure-ended dialogue. These findings show that cybersecurity assistance varies with conversational history and model deployment, motivating evaluation beyond direct-request refusal rates.

cs.AI↗
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

The shifted-prime Erdős-Wintner law for primitive-root determinant densities: extremal order, dimension zero, and Fourier decay

For a prime $p$, let $c(p)=\frac{φ(p-1)}{p-1}\prod_{j\ge1}(1-p^{-j})$, the limiting density of matrices over $\mathbb F_p$ with primitive-root determinant. Its limiting law over the primes is the classical continuous shifted-totient law on $[0,1/2]$. We prove that this law has Hausdorff dimension zero and is carried by the Liouville numbers; its lower and upper dyadic $L^q$ dimensions vanish for $q>1$. Its logarithmic image $μ_f$ under $x\mapsto-\log x$ is Rajchman and satisfies $|\widehat{μ_f}(τ)|\ll(\log|τ|)^{-1/3}(\log\log|τ|)^{1/3}$ with effective constants and threshold. For $U_T$ uniform on $[0,T]$, $(\log T)|\widehat{μ_f}(U_T)|$ converges in law to an explicit random Euler product $W_F$, and $T^{-1}\int_0^T|\widehat{μ_f}(τ)|^s dτ\sim\mathbb E[W_F^s]/(\log T)^s$ for every fixed real $s>-1$. As $h\downarrow0$, $\sup_aμ_f([a,a+h])=\mathfrak S_2e^{-γ}/\log(1/h)+O(\log^{-2}(1/h))$, where $\mathfrak S_2$ is the twin-prime singular series; every maximizing left endpoint lies within $h$ of $\log3$ for small $h$. Classical arguments give $\min_{p\le x}c(p)\sim e^{-γ}/\log\log x$; we obtain effective additive remainders and extrema in fixed reduced progressions. The laws of $σ(p-1)/(p-1)$ on $[3/2,\infty)$ and $\log(φ(p+1)/φ(p-1))$ on $\mathbb R$ also have Hausdorff dimension zero; the logarithmic divisor-sum law is Rajchman. Rational bounds for $1/c(p)$ without complete factorization of $p-1$ yield certified searches for fully splitting negacyclic number-theoretic transform primes at fixed power-of-two length. We determine the second distinct squared norm of tensor products of root lattices $A_n$ ($n\ge1$), yielding cyclotomic codifferent shell gaps and a uniform smoothing asymptotic at $ε=2^{-cφ(m)}$ for $c>2\log_2(1+\sqrt6)$. These results are unconditional.

cs.CR↗
arXiv2026

Topological Fraud Detection in Latent Transaction Spaces

Working entirely on topologically anonymized embeddings, we perform fraud detection using iterative rounds of unsupervised filtering followed by supervised sniping. The result is an ultra-low latency privacy--preserving triage that allows institutions to flag suspicious activity without compromising Personally Identifiable Information.

cs.LG↗
arXiv2026

SAEs Can Improve Unlearning: Dynamic Sparse Autoencoder Guardrails for Precision Unlearning in LLMs

Machine unlearning is a promising approach to improve LLM safety by removing unwanted knowledge from the model. However, prevailing gradient-based unlearning methods suffer from issues such as high computational costs, hyperparameter instability, poor sequential unlearning capability, vulnerability to relearning attacks, low data efficiency, and lack of interpretability. While Sparse Autoencoders are well-suited to improve these aspects by enabling targeted activation-based unlearning, prior approaches underperform gradient-based methods. This work demonstrates that, contrary to these earlier findings, SAEs can significantly improve unlearning when employed dynamically. We introduce $\textbf{Dynamic DAE Guardrails}$ (DSG), a novel method for precision unlearning that leverages principled feature selection and a dynamic classifier. Our experiments show DSG substantially outperforms leading unlearning methods, achieving superior forget-utility trade-offs. DSG addresses key drawbacks of gradient-based approaches for unlearning -- offering enhanced computational efficiency and stability, robust performance in sequential unlearning, stronger resistance to relearning attacks, better data efficiency including zero-shot settings, and more interpretable unlearning.

cs.LG↗
arXiv2026

RPP: A Certified Poisoned-Sample Detection Framework for Backdoor Attacks under Dataset Imbalance

Deep neural networks are highly susceptible to backdoor attacks, yet most defense methods to date rely on balanced data, overlooking the pervasive class imbalance in real-world scenarios that can amplify backdoor threats. This paper presents the first in-depth investigation of how the dataset imbalance amplifies backdoor vulnerability, showing that (i) the imbalance induces a majority-class bias that increases susceptibility and (ii) conventional defenses degrade significantly as the imbalance grows. To address this, we propose Randomized Probability Perturbation (RPP), a certified poisoned-sample detection framework that operates in a black-box setting using only model output probabilities. For any inspected sample, RPP determines whether the input has been backdoor-manipulated, while offering provable within-domain detectability guarantees and a probabilistic upper bound on the false positive rate. Extensive experiments on five benchmarks (MNIST, SVHN, CIFAR-10, TinyImageNet and ImageNet10) covering 10 backdoor attacks and 12 baseline defenses show that RPP achieves significantly higher detection accuracy than state-of-the-art defenses, particularly under dataset imbalance. RPP establishes a theoretical and practical foundation for defending against backdoor attacks in real-world environments with imbalanced data.

cs.CR↗
arXiv2026

The Value of Spike Timing: A Leakage-Resistant Benchmark of SNN Design Choices for Network Intrusion Detection

Spiking neural networks (SNNs) are increasingly studied for network intrusion detection, but comparative evidence on how neuron models and spike encodings affect performance remains limited. Evaluation choices can further influence the observed results when preprocessing, capture structure, or scenario information crosses the train-test boundary. We evaluate nine snnTorch neuron families with three spike encodings, yielding 27 SNN configurations across four intrusion-detection benchmarks. We first screen the complete design space and then repeat the evaluation using train-only transforms, seed-independent partitions, and capture- or scenario-aware separation. At the shared operating point, LeakyParallel/latency achieved the lowest mean rank in both sweeps but was not statistically separated from Leaky/latency; their screening order reversed when CTU-13 was excluded. Overall, screening and confirmation retained substantial rank agreement despite changes in measured performance. We then examine what latency coding contributes by holding the active spike set fixed and changing only its temporal organization. At the main T=25 operating point, mapping feature magnitude to spike time improved macro-F1 on all five confirmation protocols, while spreading the same spikes through time without that mapping could either improve or reduce performance. Latency coding thus combines informative spike timing with temporal network dynamics. Sparse input encoding does not directly imply sparse internal activity, although latency coding requires fewer fanout-weighted operations than rate coding in the evaluated models. These findings motivate separate assessment of SNN design, evaluation protocol, spike timing, and computational activity for static network-flow data.

cs.CR↗
arXiv2026

Can escalation channels redirect reward hacking toward defect disclosure?

When coding agents encounter defective test infrastructure they may reward-hack: hardcoding outputs or editing test files to pass tests they cannot legitimately satisfy, a pattern that has now appeared outside benchmarks, in a coordinated multi-agent intrusion of a major AI platform's production infrastructure. The same capability that lets an agent detect and exploit a defect could let it report one, given the right decision environment. We evaluate escalation channels, structured reporting tools available to the agent at the point of conflict, as a decision-environment intervention that both reduces reward hacking and surfaces the infrastructure defects that trigger it. A $2 \times 2$ factorial separates the contributions of an escalation tool, a standalone anti-reward-hacking policy, and their combination. Across 8 frontier models spanning 5 families, the combined intervention reduces reward hacking from 23.6% to 5.3% (mixed-effects logistic OR = 9.2, 95% CI 5.0--16.8, $p < 10^{-12}$) with no detectable cost or performance overhead, eliminating it entirely for 6 of 8 models. Escalation and hacking are near-perfectly mutually exclusive, with 98.7% of escalations involving no hacking (100% under the combined intervention). Beyond reduction, escalation channels function as diagnostic infrastructure: on top of monitoring, escalation adds +10.1 percentage points of defect detection coverage and is more accurate once it fires (99.4% vs 85.8%). Unlike containment-based approaches that risk outpacing growing model capabilities, escalation channels redirect capability toward disclosure rather than exploitation.

cs.AI↗
arXiv2026

Defense-as-Skill: Evolving Runtime Guard Skill for Skill-Augmented Agents

Skill-augmented agents load reusable skills as persistent runtime context, improving task performance but also giving malicious skills a durable channel for steering future actions. Such skills may leak secrets, corrupt code, bypass approvals, or stage data for exfiltration only after a concrete user task and workspace state make the unsafe action appear useful. This makes pre-install vetting insufficient and calls for runtime, task-conditioned protection. We propose Defense-as-Skill, a defense paradigm that implements the runtime guard itself as an installable, inspectable, and editable skill. Our guard, SkillSonar, runs alongside untrusted task skills and checks sensitive actions against the user's task boundary, routing each action to an allow, replan, or confirmation decision without modifying the underlying agent runtime. To study this setting, we construct SCOPE-R, a task-conditioned dataset covering 6 risk families and 21 sub-categories, with 206 attack-confirmed malicious instances and 43 benign tasks. We then improve SkillSonar on the SCOPE-R training subset using runtime guard-skill evolution, a Monte-Carlo Tree Search procedure that evolves the on-disk guard skill from feedback on the rollouts. Across Claude Code and OpenClaw, the evolved guard substantially reduces attack success while maintaining a favorable safety-utility trade-off; on repeated GLM-5 runs, ID ASR drops from 0.482 to 0.104 and OOD ASR from 0.606 to 0.115, with gains transferring across victim models, held-out risk families, and external benchmarks. These results show that runtime guards can be skill-native, transparent, and evolvable policy artifacts for skill-augmented agents.

cs.CR↗
arXiv2026

ICS-Sniper: A Targeted Blackhole Attack on Encrypted ICS Traffic

Modern industrial control systems (ICS) increasingly host their Supervisory Control and Data Acquisition (SCADA) services in the cloud to reduce the costs of large-scale automation. To protect site-SCADA communications, ICS operators commonly use VPN tunneling and standard security practices. We show that, despite these security measures, an on-path Internet adversary can disrupt ICS operations without infiltrating the ICS perimeter, breaking encryption, or knowledge of the control logic. We present ICS-Sniper, a targeted blackhole attack that analyzes the VPN traffic metadata (sizes, direction, timing of packets) to identify narrow time windows, called critical superperiods, during which the site-SCADA traffic would likely contain highly critical commands or data. Post-analysis, in a subsequent operational cycle, ICS-Sniper drops a small set of payload-carrying packets in the critical superperiods to disrupt the ICS's operations. We demonstrate three attacks on two realistic modern Secure Water Treatment (SWaT) plant testbeds that can potentially violate the operational safety of the ICS while evading state-of-the-art ICS attack detectors.

cs.CR↗
arXiv2026

The Implications of Linguistic Illegibility for LLM Security

LLMs are trained to generate natural language. However, various strands of evidence indicate that an LLM's externalized linguistic outputs and mechanistically-extracted linguistic features can be an unreliable lens for understanding internal model computation. We introduce the term ``linguistic illegibility'' to broadly refer to scenarios in which an LLM's externalized or mechanistically-probed language artifacts fail to represent how the model actually thinks. We argue that the specter of linguistic illegibility is unavoidable for LLMs whose internal computations are not directly expressed via language, but rather math over activation spaces (with lossy translations between activation spaces and natural language happening at the bookends). If linguistic illegibility is always possible, then security mechanisms that rely on a model's linguistic self-reporting (e.g., chain-of-thought monitoring, constitutional self-critique, activation probing for linguistically-defined feature vectors) can never be completely sound; the model sandbox will always need isolation techniques whose guarantees do not depend on reading a model's linguistic state at all. We argue that observing a model's outputs using taint tracking is a promising approach for an effective sandbox: regardless of how a model linguistically self-reports, a taint tracking policy can define, a priori, various pieces of system state that should never be influenced by model-produced data. We also discuss several additional sandboxing mechanisms (e.g., robust virtualization, third-party auditing of sandboxing configurations) which collectively provide a critical floor beneath linguistic monitoring, and would have mitigated recent sandbox exploits by frontier models.

cs.LG↗
arXiv2026

Cost-Aware Hierarchical Multi-Agent Ransomware Detection and Family Attribution under Analysis Budgets

Sandbox execution and memory forensics are among the most constrained resources in malware triage. Static analysis can scale to millions of files, whereas dynamic and memory analysis require minutes of analyst controlled infrastructure for each sample. Despite this difference, multimodal ransomware detectors often apply every modality to every sample, causing analysis cost and time to verdict to increase linearly with sample volume even when static evidence is already sufficient for a decision. We present a cost aware Hierarchical Multi-Agent System that formulates evidence acquisition as a budgeted sequential decision problem. Specialist agents generate schema validated risk signals for each modality, domain controllers aggregate these signals, and a Meta-Orchestrator begins with static evidence and escalates to dynamic and memory evidence only when confidence is insufficient or agents within a controller disagree. An optional, bounded, locally hosted large language model reviewer can adjust a verdict by at most one tier but cannot replace the deterministic pipeline. Each decision is recorded with a complete provenance trace. In multiple runs over 12439 samples from 16 ransomware families and benign samples, the deterministic HMAS achieves F1 0.93 and macro F1 0.97, resolving 57.95% of cases using static evidence alone, 35.83% after adding dynamic evidence, and only 6.21% through the full pipeline. The average internal analysis cost is 6.65 units, compared with 12 for exhaustive analysis, representing a 44.6% reduction. Standalone leave-one-family-out testing further shows that accuracy on families held out during tuning falls to 0.26 to 0.64 outside the Benign and high support classes. We report these results alongside a cost sensitivity analysis, a partial leave one component out ablation, and a full scale comparison with learned early and late fusion and cascade baselines.

cs.CR↗
arXiv2026

Capability-Gated Language Models: Security Composes, Utility Does Not

Deployed language model safeguards (safety fine-tuning, filtering, unlearning) vary by principal only outside the model weights: filters are reconfigured, tiers are multiplied, and artefacts are reissued; inside one set of weights every request meets the same model configuration. This motivates us to define capability-gated deployment: per-principal access control inside one set of weights, whose configurations form a lattice - meets accumulate a principal's restrictions and joins pool a coalition's reach. We instantiate it by sparse rank gating over an existing nested-factorisation mechanism, guide profile search with one-pass attribution, and read every result once from a pre-registered held-out split. Security approximately composes: provably exactly at meets under a monotone-elicitation assumption we falsify pointwise. In two lineages the median held-out meet deepens suppression; the one effect surviving correction strengthens it. Utility does not: individually harmless profiles can compose to retention and fluency damage, and no compositional bound exists.

cs.CR↗
arXiv2026

Large Language Models for Agentic NetOps and AIOps: Architectures, Evaluation, and Safety

Large language models (LLMs) are increasingly being used in network operations (NetOps) and artificial intelligence for IT operations (AIOps) for tasks ranging from telemetry retrieval and incident diagnosis to configuration planning and bounded remediation. As these systems acquire greater access to operational tools, a central question is whether operational assurance develops commensurately with the authority granted to them. This survey examines that question through a structured, evidence-stratified review of agentic NetOps and AIOps, spanning directly evaluated LLM systems, operational mechanisms, transferred evidence, and normative guidance. We organise the field around autonomy, tool scope, evidence traces, assurance controls, evaluation, security, and governance, and introduce an operational assurance contract that links each autonomy level to permitted tools, required evidence, independent gates, execution budgets, rollout and rollback duties, and audit requirements. The synthesis indicates a capability-assurance gap: evidence is comparatively strong for read-oriented assistance and tool-grounded diagnosis, but becomes substantially less complete as systems approach configuration change, bounded execution, and closed-loop operation. We therefore develop a workflow-level evaluation framework covering evidence quality, tool use, policy and invariant compliance, staged execution, recovery, calibration, cost, and human intervention, together with a threat model for prompt-borne attacks, compromised operational evidence, excessive agency, privilege boundaries, and auditability. The resulting view treats agentic NetOps and AIOps as constrained operational control, in which increasing autonomy requires correspondingly stronger evidence, independently enforced safeguards, and recovery mechanisms.

cs.NI↗
arXiv2026

Condense to Conduct and Conduct to Condense

In this paper, we present the first explicit examples of low-conductance permutations. The notion of conductance of permutations was introduced by Dodis et al. in "Indifferentiability of Confusion-Diffusion Networks", where the search for low-conductance permutations was first initiated and motivated. As part of our contribution, we not only provide these examples, but also offer a general characterization of the problem: we show that low-conductance permutations are equivalent to permutations possessing the information-theoretic properties of Multi-Source-Somewhere-Condensers, a specific variant of somewhere condensers.

cs.CR↗
arXiv2026

SIR: Self-improving Red-teaming for Compute Use Agents

Computer-use agents (CUAs) are agents powered by vision-language models (VLMs) that perceive a screen and operate an operating system through mouse, keyboard, and terminal interactions to automate everyday digital tasks. Their exposure to untrusted content creates a risk of indirect prompt injection (IPI), where an adversary embeds instructions in content the agent reads to redirect it toward actions that violate the user's intent. Evaluations based on fixed, hand-written injections may underestimate the risk posed by adaptive adversaries. We present SIR, a black-box self-improving IPI framework that (i) composes task-specific injections from a small library of reusable red-teaming principles stated in plain language and (ii) uses an iterative feedback loop to analyze unsuccessful attack trajectories and distill new, named principles that are retained in a shared library and reused across tasks. We target operating-system-level compromise and evaluate outcomes through deterministic checks on filesystem, service, and permission state rather than an LLM judge. An attack counts as successful only when both the adversarial objective and the benign user task are completed in the same execution. We evaluate three frontier CUAs, allowing SIR up to 10 attack attempts per case. It achieves joint attack success rates of 24% on Claude Opus 4.8 and 28% on Gemini 3.5 Flash, compared with 4% and 0%, respectively, for the benchmark's fixed, hand-written injection. The red-teaming principles discovered against one victim also improve attacks against other victims, including a different model family, without additional feedback.

cs.CR↗
arXiv2026

Spruce: Scalable Private Outsourced Retrieval Using Compact Embeddings

Retrieval-Augmented Generation (RAG) has made dense retrieval over large document collections a standard building block. Organizations increasingly outsource vector indexes to untrusted clouds, exposing proprietary corpora and user queries. Cryptographic protection is challenging because each query searches corpus-scale state, causing computation, correlated randomness, and communication to grow with the corpus. At million-document scale, a naive secure implementation takes minutes and about 90 GB of communication per query. Even recent optimized systems require 10--22 seconds. We propose Spruce (Scalable Private Outsourced Retrieval Using Compact Embeddings), which co-designs representations with the cryptographic protocol. Spruce learns compact binary codes that preserve candidates for full-precision reranking, replacing corpus-wide embedding scoring with efficient Hamming-distance computation under two-server multi-party computation (MPC). A corpus-calibrated fixed-radius protocol avoids multi-round candidate selection while preserving retrieval quality. Spruce also provides private cluster pruning, which trades minor quality loss for substantially less computation, and a one-core owner-operated dealer that removes cloud OT preprocessing bottlenecks. Across four corpora containing 383K--5.42M documents, Spruce preserves the original search quality with median candidate sets of only 382--1,952. At 10 Gbps inter-server bandwidth, full scans take 0.21--2.97 seconds, $4.8$--$6.7\times$ faster than the closest measured prior work. Private pruning takes 0.06--1.09 seconds, achieves $13.1$--$22.9\times$ speedups, and retains $93.9\%$--$97.3\%$ of full-float NDCG. On the largest corpus, pruning and the dealer jointly improve sustained throughput by $31.5\times$ at 1 Gbps per link.

cs.CR↗
arXiv2026

Evaluating a Layered Prompt-Injection Defence for the Model Context Protocol: A Record-Level Audit of Decision Conventions, Corpus Provenance and Reproducibility

The Model Context Protocol (MCP) extends the prompt-injection attack surface of large language model applications to tool descriptions, parameter schemas and tool outputs. Defences for it report detection figures that are not comparable, because each is measured on its authors' own corpus under a decision convention that is rarely stated. This paper audits one such evaluation at the level of individual decision records. Its object is CASCADE, a fully local layered defence (rule matching, embedding similarity and an optional local language-model review), run in three configurations on a frozen 5,000-sample corpus under a pinned code revision; the corpus, the per-sample records and the analysis scripts are public. Four findings result. First, the convention that collapses allow, review and block onto a binary label sets the headline: the pipeline without review reports an 11.70% false-positive rate when referrals count as positives and 1.51% when only denials do, while referring 68.5% of all traffic to a human. Second, the construction of the corpus shapes the aggregate: 65.6% of records are texts placed in one of five fixed wrappers, the wrapper alone identifies the label, and for the same text wrapping lowers the false-positive rate from 21.2% to 3.2% and raises detection from 86.0% to 98.0%. Third, the published description does not identify what ran: the rule module inside the pipeline is not the rule engine evaluated alone; the block threshold produced denials only among the first 48 requests; and, without review, every later denial, including all 23 denials of benign requests, came from an output guard run on an echo of the input. Fourth, the local review model, invoked on 32.6% of requests at 2.5 s each, changes no binary outcome but converts 1,492 referrals into denials. A ten-item reporting checklist is derived from these findings.

cs.CR↗