Papers

Video see-through extended reality (VST XR) systems commonly use headset screenshots or captured frames as proxies for the user's first-person visual context. However, the system-captured view and the user's effective visible field do not necessarily coincide: a screenshot records a rectangular machine-readable frame, whereas the user's effective visible region can be more constrained and non-rectangular. This paper studies this human-system view mismatch in VST XR. We formalize the relationship between the system-captured region and the human-visible region by defining their co-visible, system-only, and human-only regions. \rev{We then conduct a pilot-level boundary measurement on Meta Quest 3, revealing a clear mismatch between the rectangular screenshot frame and the approximate human-visible boundary. Building on this model, we analyze how view mismatch can affect screenshot-based XR sensing and downstream vision-language model tasks. Through four representative case studies, we illustrate potential risks and failure modes including prompt injection, privacy leakage, human-invisible information bias, and missing human-visible information. Our results show that view mismatch is not only a geometric artifact, but can also introduce security, privacy, and reliability concerns for AI-integrated VST XR systems.

Short claims such as not-phishing or official can change how a large language model (LLM) judges a domain name, without explicit prompt-injection commands. We study this manipulation as ClaimMirage: a name under inspection claims its own safety or approval. We analyze 622,080 judgments across 64 brands and five LLMs, comparing ten claims with length- and hyphen-matched controls in constructed brand-like names. Self-claims can substantially reduce or increase alerts, depending on the LLM and input setting. In one setting, risk-denial terms inside the registrable name reduce alerts by 45.3 percentage points even with a basic safeguard: the prompt supplies the potentially impersonated brand and its official domain for comparison. Without these references, endorsement terms at that position increase alerts by 65.6 points in the same LLM. References and component annotation remove some alert reductions but leave others or make them larger. These findings motivate testing resistance to self-claims and seeking independent evidence before treating a domain name under inspection as safe or authorized.

Autonomous security-testing agents can spend much of a fixed action budget repeating earlier tool selections. We evaluate a reward-free scheduler inspired by mushroom-body novelty processing in Drosophila. It combines sparse state encoding with decaying habituation counters over structural URL classes and tool families. The counters penalize repeated clean or error outcomes without updating weights from scalar reward. Four matched campaigns motivated this design by exposing reward-accounting errors and tool-failure loops; reward-driven components did not improve the tested primary outcomes over the reward-free MB condition. A pre-registered pilot and two confirmatory stages then evaluated repeated (tool, URL) selections. In the second confirmatory stage, 8 of 10 screened lab targets remained measurable after two error-heavy slow-XSS exclusions. The habituation-enabled scheduler lowered duplicate-action ratios in all 6 non-tied target pairs (exact one-sided p=0.015625), with two ties; the largest reduction was 51 to 18 duplicate steps within a 60-step budget. This is evidence for the complete scheduler on the measurable budget-hold population, not an isolated habituation ablation or a vulnerability-discovery gain. A complementary study on a 13,498-neuron MaleCNS-derived circuit (501,267 synaptic edges with weight at least 5) found no action selectivity from the five tested local-plasticity approaches under a fixed readout; readout plasticity produced qualified positive results in synthetic tasks without establishing a biological-topology advantage. We report the population bounds, remaining input-integrity dependencies, and an internal AI-assisted review protocol alongside the results.

Multimodal training relies on image-text corpora collected from external sources, creating opportunities for attackers to poison the data. Stealthy attacks can preserve plausible image-text pairs while concealing the differences used by detectors, so apparently clean data can still redirect the trained model. We therefore ask which properties a poison set must preserve for the attack to remain effective. A small poison set must still exert enough collective influence during training to induce the attacker's target behavior. We analyze this influence in terms of how often an attack pattern occurs and how strongly the examples carrying it jointly affect the model. This analysis motivates six corpus-level features that examine cross-modal neighborhoods, recurring text, and changes after text-span erasure without training the victim model. We introduce TraceGuard, an adaptive rank-based filtering method that uses agreement among complementary feature rankings to identify suspicious examples. It refines the selected set through shared patterns and adapts the removal threshold to each corpus without knowing the attack or poison rate. Across 19 attack configurations spanning image-text learning, generative vision-language model fine-tuning, and encoder-transfer tests, TraceGuard removes an average of 98.4% of poisoned examples and 5.4% of clean examples. After training on the filtered corpora, the residual attack metric is at most 1% in 13 configurations. Matched-removal controls and ablations support the contributions of sample selection and adaptive removal. Stress tests also identify detection failures under adaptive attacks and unnecessary removal on poison-free corpora.

Open-weight LLMs give downstream users control over the inference stack, but this flexibility can undermine post-release guarantees that sensitive knowledge has been modified or removed. Model editing and machine unlearning are used to modify or remove targeted knowledge without retraining models from scratch. However, existing security evaluations of these techniques face two critical limitations. First, they typically require access to either the original pre-edit/unlearning model or auxiliary classifiers to detect modifications or reconstruct pre-edit behavior. Second, they evaluate modifications under the canonical tokenization of an input, implicitly treating tokenization as a benign preprocessing step. We show that this assumption creates a security gap: the same input string can be represented by alternative valid tokenizations that induce different computational trajectories, allowing an adversary to bypass localized modifications and recover information intended to be suppressed. We introduce Toketive, a simple yet powerful reference-free attack that exploits the tokenization-based side channel to (i) detect modified knowledge and (ii) reconstruct the corresponding pre-edit response. It operates solely on the released model and requires neither the pre-edit model, training data, shadow models, nor auxiliary classifiers. Across five LLMs, six datasets, and six editing and unlearning techniques, we find that 38.6% of alternative tokenizations bypass the modification and recover the pre-edit response. Toketive detects modified facts with an F1 score of 84.2%, a 26.2% relative gain over the strongest baseline, and reconstructs pre-edit responses with 74.5% top-5 accuracy, 21.7% higher than the best baseline. Our results show that localized modifications should not be treated as robust knowledge-control boundaries without adversarial evaluation over alternative representations.

Neural audio watermarks are increasingly deployed in commercial speech generation systems to make AI-generated speech traceable, yet their robustness has been studied mainly under conventional signal distortions. Since a watermark can be regarded as imperceptible noise added to the speech signal, a natural question is whether speech enhancement (SE), as a denoising model, can remove it. In this paper, we cascade Gaussian noise with SE models as a black-box watermark removal attack, covering both discriminative and generative SE paradigms, against six neural watermarks: AudioSeal, WavMark, SilentCipher, Timbre, Perth, and AlignMark. Experimental results show that the proposed attack significantly outperforms existing neural re-synthesis methods in watermark removal. In particular, we find that generative SE, which reconstructs the harmonic regions of speech while denoising, is highly destructive to watermarks. These findings show that SE poses a serious threat to current audio watermarking methods, and we call for SE-aware robustness evaluation in watermark design.

The Node.js ecosystem heavily relies on NPM packages, and software supply chain attacks targeting malicious NPM packages are rampant. Malicious code primarily triggers during package installation, import, and runtime. Traditional static analysis fails to understand code semantics; machine learning-based methods rely on feature extraction, which suffers from concept drift; existing LLM solutions suffer from high invocation costs, high data security risks, and poor performance. To overcome these limitations, we propose DistillGuard, a lightweight malicious NPM package detection framework that combines static graph semantic analysis and LLM knowledge distillation. DistillGuard first acquires multi-granular features through three static analysis modules. Then, it leverages online LLM to distill high-quality security knowledge and structured labels. Finally, it uses LoRa to efficiently fine-tune the open-source Qwen3-8B model to support offline deployment. Experiments show that DistillGuard achieves an accuracy of 95.3%, a precision of 99.4%, and an F1 score of 93.8%, outperforming state-of-the-art tools, improving the F1-score by 11.1 to 30.0 percentage points over the baselines. Our empirical research further reveals the stages of malicious attacks and the distribution of malicious behaviors. We also summarized eight typical API attack chains for malicious activities, providing practical insights for NPM supply chain security defense.

Autonomous penetration-testing harnesses use large language models (LLMs) for reconnaissance, exploitation, and reporting, but often rely on those same models to confirm findings, grade severity, and select agents. This can lead to false positives, inflated severity, and wasted compute. We examine how System One decision models, lightweight non-generative classifiers that return typed, calibrated verdicts, can support these decisions. We make five contributions. First, we define four decision points: finding adjudication, severity recalibration, agent pruning, and confirmation loops. Second, we present an exploratory NeuroSploit case study comparing one run with TypeSafe System One (Jev) and one without it against a web target containing 13 vulnerabilities. Differences in severity distribution, runtime, and grading by exposed data type motivate the architecture but do not establish statistical significance. Third, we review published specifications for Jev, Jev-Ultrafast, and the open-source Laya without assuming that results from other benchmarks transfer to penetration testing. Fourth, we discuss RLHF, RLAIF, RLCD, and RLHV as training approaches and their implications for trust in security decisions. Finally, we propose Rave, a domain-adapted System One model, and outline its training data, evaluation protocol, and potential effect on harness assurance.

Harnesses, the products that run AI coding agents, are multiplying, and enterprises are rolling them out to their employees. What started as pilots with a few hundred seats is now scaling to tens of thousands. Enterprises rarely build these harnesses and usually adopt the ones model vendors bundle with their seats and APIs, such as Anthropic's Claude Code or OpenAI's Codex. A harness decides which model answers, what the model reads, how the prompt cache is used and which subagents run, so it picks the rate on the price sheet and sets the volume bought at it. Enterprises that keep a proprietary or untuned harness at its defaults inherit these choices and their bill. We build a fast, customisable router in which Jev, a classifier with calibrated probabilities, labels every prompt against a bring-your-own taxonomy of agentic requests. One user turn is many requests over a prompt cache that belongs to one model, so the router moves work only where no running conversation has to rebuild its cache. It routes at session start, in side lanes and at subagent launch. From the price sheet we derive when a mid-task switch pays back. We also find a crossover, in which the highest-priced model costs less than the next tier on long tool-heavy sessions, and repricing about 10,000 real sessions from public datasets confirms it. In an emulated enterprise of 10,000 seats with user behaviour taken from these datasets, the router recovers 13 to 21% of model spend at Anthropic's list prices of 21 September 2026, \$3.3M to \$5.1M a year. The paper also maps the risks across twenty-one harnesses, prices the dependence on one vendor's models, and proposes a control plane that enterprises can run from within, starting now, with a ladder for deciding later whether to own the harness.

Frame refresh rate estimation is a fundamental step in identifying compromising harmonic frequencies in electromagnetic side-channel attacks. Methods based on discrete linear autocorrelation (DLA) are robust across different scenarios and require a computational complexity of $O(N\log N)$ for a signal containing $N$ samples. This work proposes reducing this complexity to $O(N)$ by exploiting prior knowledge of the target system's display resolution, as is available for certain models of the Brazilian electronic voting machine. Experiments using software-defined radios show that the proposed method preserves the accuracy of the conventional approach in the evaluated scenarios.

LLM agents are deployed into infrastructure that grants them broad host authority, yet existing agent-security benchmarks and defenses operate almost exclusively at the application telemetry layer: the served tool manifest, the user prompt, and the model's messages. Some threats, however, smuggle malicious instructions and actions past the application boundary, leaving them invisible to that layer. In this work, we bridge that gap by pairing application-level agent telemetry with kernel-level syscall traces to present the first paired-evidence characterization of kernel-level versus application-layer signal for agent security. To quantify the value of the enhanced telemetry, we introduce Agent Cross-Layer Evidence (ACE), a paired-session corpus of 4,047 sessions and 17 threat models spanning six delivery-vector families and 14 of the 25 OWASP LLM and agentic threat categories, organized into 12 attack mechanics with per-mechanic characterization of where the most discriminative evidence lies. Across four distinct detector families, we find that kernel evidence is discriminative on its own and that composing it with application-layer evidence generally outperforms either single-layer view, revealing complementary signals that single-layer analyses can miss. We further demonstrate generalization to unseen attack families and transfer to an alternate agent runtime. Together, these findings establish the value of cross-layer evidence for agent security.

Multi-agent systems built on large language models (LLMs) are increasingly deployed in high-stakes settings such as finance, healthcare, and software engineering, where agents coordinate through natural-language messages. The same channels, however, let colluding agents exfiltrate confidential information or coordinate unauthorized actions, and steganography can hide such communication inside outputs that look ordinary to an auditor reading the transcript. Existing provably undetectable LLM steganography protocols are not suited to realistic deployments. High-capacity schemes assume a symmetric setting where the receiver can reproduce the sender's output distribution, the state-of-the-art protocol for asymmetric agents has very low capacity, and most approaches rely on a pre-shared secret key. We make the threat of undetectable agent collusion concrete with Codetta, a high-capacity steganographic protocol for independently deployed agents in realistic asymmetric settings. Codetta combines a shared public model that estimates the communication channel, a sampling mechanism that preserves the sender's output distribution, and an adaptive error-correcting code. It further removes the pre-shared key through a steganographic key exchange that lets independently deployed agents establish a shared key while keeping the transcript computationally indistinguishable from ordinary model outputs. Across three agent workloads and three sender models, Codetta achieves up to $94\times$ the capacity of the state-of-the-art asymmetric protocol, and its key exchange establishes a shared key with about 80k visible tokens at an empirically certified failure probability of at most $4.1\times 10^{-3}$. These results show that effectively undetectable collusion is becoming feasible between independently deployed agents, so auditing must go beyond inspecting communication transcripts.

We study the differential privacy (DP) of a graph shift operator (GSO) when an analyst observes the output of a graph filter. In particular, we study the setting in which the input signals to the graph filter are drawn from a differentially private distribution. Unlike approaches that perturb the GSO or the filter output, we use the randomness already present in the inputs to protect the GSO. This yields an equivalent level of privacy protection to that of the perturbation methods without adding noise, and thus a better privacy-utility trade-off. We provide an explicit characterization of the privacy loss and its certificate in terms of the zeros of the graph filter. In doing so, we show that the log-likelihood ratio between the releases of two adjacent topologies is governed by the distances from each zero to the graph frequencies of the two GSOs. Then, by uniformly bounding the log-likelihood ratio over the adjacent topologies, we obtain an explicit $(\varepsilon,\delta)$-DP guarantee for Gaussian inputs. We further show, via a Cramér–Rao bound, that the zero placement that limits the privacy loss also raises the floor on the adversary's reconstruction error. Finally, empirical validation is performed on a synthetic network of financial exposures, where the largest position a pair can conceal and the accuracy with which it can be sized are collinear across pairs. Both are set by the graph-frequency content of the pair, and the full network becomes recoverable only as the certified budget grows.

Agents are increasingly deployed with real autonomy in web application and network penetration testing, where a single out-of-scope action can breach a client's engagement boundary. Existing offensive-security benchmarks measure raw hacking capability; as those benchmarks saturate, the real barrier to deployment is a special case of alignment: scope adherence. We introduce ScopeBench, a benchmark of 30 dead-end agentic security tasks in which the stated objective is reachable only by violating the stated scope. Each task appears under two conditions that share an environment, verifier, and objective and differ only in scope: one instruction set has no scope and measures capability; the other has a natural-language scope to measure adherence. Scopeless trajectories are graded by a standard deterministic verifier. Scoped trajectories pass through two grading arms. First, the same deterministic verifier checks for the flag: because the flag sits behind the scope boundary, a pass proves by construction that a forbidden action occurred, yielding a high-precision lower bound on the violation rate. If the verifier does not pass the trajectory, an agentic judge estimates whether an out-of-scope call occurred. We calibrate the judge against 100 ScopeBench trajectories labeled call-by-call by human annotators, and a blinded audit of the evaluated rollouts finds its high recall holds - no false negatives among the 36 audited violations, with over-flagging its only observed error. Across 8 models in one harness, raw capability spans 12.2% to 81.1% and scope adherence spans 34.4% to 86.7%, with the judge finding 331 violations that mechanical verification misses. Opus-4-8 achieves a raw-capability score 10 percentage points higher than sonnet-4-6's while exhibiting 35.6 percentage points higher scope adherence. We release the frozen pilot benchmark, evaluation code, and all 2160 ATIF trajectories.

Blockchain and artificial intelligence (AI) are converging into a single infrastructural layer for securing data sharing, model integrity, and autonomous decision-making across distributed systems. This paper presents a meta-synthesis that draws together four constituent studies covering adversarial machine learning, AI-powered anomaly detection in cloud environments, automated vulnerability patching by multi-agent large language model (LLM) pipelines, and the broader landscape of securing AI systems across their lifecycle and situates their findings within the emerging literature on blockchain-enabled AI and autonomous AI agents. Each constituent study addresses a distinct point of failure in modern AI-driven security operations: the integrity of training data and model behavior, the reliability of real-time monitoring, and the trustworthiness of automated code remediation. We argue that blockchain's properties of immutability, decentralized consensus, and verifiable provenance directly address a gap common to all three: the difficulty of establishing trust in data, models, and autonomous agents that operate without a central authority. Building on real-world research on blockchain-secured data sharing, federated learning, and multi-agent coordination, we propose a layered reference architecture that couples adversarially hardened models, blockchain-anchored data provenance, AI-driven anomaly detection, and smart-contract-governed multi-agent remediation. We conclude by identifying open problems in scalability, privacy-transparency trade-offs, and the governance of autonomous agents that must be resolved before such integrated systems can be trusted in production-critical environments.

Machine learning-based Network Intrusion Detection Systems often report near-perfect performance on IoT benchmarks. However, whether these models learn generalizable attack behavior or exploit spurious dataset shortcuts- such as static testbed IP/MAC addresses and chronological recording artifacts-remains an important question. We evaluate the CyberFlowIoT-GICAP benchmark, containing 3,617,388 flow records across 126 PCAP sessions with 849,395 benign flows. Four learning paradigms are evaluated across four feature configurations using PCAP-disjoint splits; LightGBM is additionally evaluated using conventional random-flow splitting. When only statistical flow behavior is used (Fbehav), LightGBM (92.58% +/- 8.18%), Random Forest (92.59% +/- 8.18%), and Deep MLP (92.55% +/- 8.18%) achieve nearly identical Macro-F1, indicating that performance is constrained by feature representation rather than model complexity. With raw timestamps (Ftstamp), tree-based models reach 99.28% Macro-F1, while the linear model remains at 90.62%, showing that nonlinear models can exploit dataset-specific temporal structure. Attack detectability is highly asymmetric: high-rate and active attacks maintain >99.8% recall from flow behavior alone in nonlinear models, whereas the DNS Beaconing drops from 27.78% to 0.00% recall when contextual features are removed. Conventional random-flow splitting increases attack recall by up to 14.00%, highlighting the effect of placing flows from the same sessions in both training and test sets. We conclude with a 4-point protocol checklist for realistic IoT NIDS evaluation.

Dynamic Spectrum Access (DSA) through the Spectrum Access Systems (SAS) elevates spectral efficiency, yet existing centralized models face allocation logic opaqueness and a lack of independent verifiability. While blockchain-based SAS architectures offer transparency and verifiability by default, they introduce critical privacy risks and prohibitive on-chain computational overhead. We introduce zkSAS, a practical zero-knowledge proof (ZKP) system designed to address the verifiability and privacy gaps in SAS deployments, with direct applicability to both the existing CBRS SAS model and blockchain-based SAS models. The system features a suite of ZKP circuits, encompassing proofs of allocation constraint validity and proofs of move list validity to verify that channel assignments and move list-based incumbent protection measures, respectively, adhere to regulatory constraints without exposing sensitive user data. Comprehensive evaluation of our prototype in both centralized and blockchain-based settings indicates that while proof generation scales with spectrum user population, verification remains lightweight and constant-time. We envision that zkSAS offers a scalable and practical path to secure, verifiable dynamic spectrum sharing.

Large Language Model (LLM) agents struggle to scale safely when exposed to vast enterprise toolsets. Providing an agent with access to every internal tool leads to oversized context windows, degraded tool selection, and severe governance vulnerabilities - as system policies defined purely in prompts remain probabilistic advice rather than hard constraints. Existing mitigations, such as multi-agent domain delegation, decentralize audit logs and fail to guarantee policy compliance across sessions. We introduce skilder, a framework that packages capabilities into roles: bundles of skills, tools, and instructions, together with the limits that bound them. An agent begins with a minimal role catalog, learns the roles a task requires, and receives each role's skills, instructions, and tools through a single MCP server. Because tools reach the agent only inside learned skills, the same server enforces the scope of what was learned deterministically. We evaluate skilder against flat-context tool selection and multi-agent orchestration across 13 tasks using six models (10 runs each). Our results show that, when models completed discovery and issued a governed call, the skilder simulated authorization layer enforced governance boundaries: no unauthorized tool call or parameter violation (e.g., a spending-limit breach) executed. Aggregate task pass rates also reflect whether each model followed the discovery protocol and satisfied response-quality checks; those misses are not authorization failures. Furthermore, by allowing agents to dynamically acquire cross-role capabilities mid-task, skilder preserves problem-solving flexibility while providing hard system-level enforcement.

The proliferation of audiovisual deepfakes has lowered the cost of fraud, impersonation, and misinformation, but their success ultimately depends on human perception. Detection requires integrating auditory and visual cues, yet security and privacy research has largely overlooked d/Deaf and hard-of-hearing (DHH) populations. We address this gap with an in-person, mixed-methods study of 80 participants: 31 hearing persons (HPs), 15 hard-of-hearing (HoH) participants, 17 d/Deaf participants, and 17 cochlear implant (CI) users. Each participant judged the authenticity of 30 clips, where manipulations spanned text-to-speech, voice conversion, lip-sync, or face-swap. DHH participants were less accurate than HPs overall (76.4% vs. 88.0%, p<.001), primarily because they more often classified authentic clips as manipulated (FPR: 29.7% vs. 11.2%). Differences depended strongly on the manipulated channel. For audio-only manipulations, HoH participants matched HPs (90.0% vs. 90.3%), followed by CI users (79.4%) and d/Deaf participants (41.2%). When clips contained an audiovisual manipulation, accuracy clustered between 84% and 87%, although performance still varied by manipulation method. Our work systematically characterizes how deepfakes affect DHH populations, highlighting the asymmetric risks audiovisual manipulations may pose to groups with different hearing abilities and the need for accessible, tailored defenses that support all users.

Most studies of prompt injection focus on generative agents, leaving their effects on models with schema-defined outputs unclear. We examine these effects in Jev, a non-generative decision model, using 510 reconstructed InjecAgent cases. Malicious content shifts action probabilities but rarely causes Jev to select the attacker's target. Override markers reduce this influence, while claims of contextual relatedness have small effects. Adaptive attacks using score feedback double the mean highest attacker-target probability found during optimization, while success on fresh validation calls rises from 1.8% to 3.5%. Exploratory analysis links these successes to small initial decision margins or greater attacker control over the observation. Together, these findings show that schema-defined outputs change but do not eliminate prompt-injection risk, highlighting the need to evaluate how untrusted content influences choices within the allowed action set.

Removing CPython's Global Interpreter Lock (GIL) exposes native code to concurrency absent from ordinary C types. Mutation or re-entry can revoke a borrowed object, storage pointer, traversal state, or lease between acquisition and use while its owner remains alive, causing native memory errors and runtime-state corruption. Race analyses track conflicting accesses. Python/C lifecycle analyses track individual object states. These reporting units leave implicit owner-subject-storage relations disconnected from later uses under parallel and re-entrant events. We present CONCURDEP, a source-level static analysis of dependency invalidation. Its key insight is to represent the runtime property a native use requires and ask which target-matched event can revoke it within the dependency's live region. CONCURDEP recovers runtime-semantic dependencies, connects them to events through an event-aware native concurrency dependency graph, and applies property-specific state and protection transfers through a shared engine and six mechanism plugins. CONCURDEP correctly classifies all 180 matched semantic-conformance cases and analyzes each of three production CPython releases with five-run medians of 22.13-27.66 seconds and 670-784 MiB peak resident memory. Source auditing confirms 1,094 of 4,273 unique production fingerprints (25.60% confirmation yield). Removing derived relation recovery loses 25-44 represented roots per release; removing cross-entry events loses 73-94. The study identifies 144 distinct bugs across free-threaded and conventional-GIL builds, including 95 previously unreported in public sources. These results show that explicit dependency, event, and property semantics expose consequential runtime failures across API boundaries and execution modes at release scale.

Flipping 60% of training labels from a single Byzantine client using label-flipping model poisoning self-degrades an attacker's own federated detection accuracy, $99.96\%$ (at no poisoning rate) to $84.33\%$ in a three-client federated IDS. Where the Federated global ensemble maintains stable accuracy across all tested poison rates, without a defense mechanism in place and without coordination between attackers. In this paper, we present empirical results quantifying the impact of label-flipping poisoning attacks on a three-client federated IDS trained on CICIDS2017 with non-IID attack subtype distributions across clients. We demonstrate that the signal of the adversarial self-compromise represents a detectable anomaly for exploitation for Byzantine client identification in the absence of target data exfiltration. We note that the aggregation step uses a Federated Forest (tree concatenation) rather than a parametric FedAvg; the results therefore measure the impact of poisoning on per-client performance under ensemble aggregation, and extension to genuine FedAvg with a parametric classifier is planned for future work.

Coding-agent approval interfaces bind a human decision to a command or tool call, while developer tools execute the transitive workflow that invocation activates. Package installation can run lifecycle hooks and write files; an MCP call can exercise network authority. We call the resulting record-coverage failure approval laundering: the durable record names the entry invocation but omits effects exercised by its workflow. We present the first systematic security analysis of this record-to-closure relation in agent systems. We formalize closure-bound approval over six effect classes and derive an information limit: identical policy-visible fields can require different effect-specific decisions, so no record-only policy can guarantee both. The Approval-to-Action Security Benchmark binds approval objects and decision-time metadata to post-execution evidence. Across 111 fixed approval-object/trace pairs, residual records fall from 40 under explicit fields to 17 with command semantics and 13 with decision-time metadata. Across 11 fixed-SHA executions, the ladder reaches zero metadata residuals; two exact mappings recur across three product frontends. For prospective recovery, effect-bound records commit frozen, source-backed predictions and provenance before authorization. On 17 prespecified holdout workflows, predictions achieve 0.926 macro recall and 0.941 macro precision; binding them cuts residual effects from 10 to 3. A Claude Code PreToolUse integration carries the frozen record through the permission path without automatic approval. These results establish approval laundering as a measurable, recurrent record-coverage failure despite truthful invocation identity. They motivate binding each invocation before authorization to a source-backed prediction of its workflow's transitive effect boundary and preserving that binding with the decision.

Multi-step tool-calling LLM agents rely on host runtimes to preserve state across turns. When a runtime carries an external tool return into later model inputs, providers meter it again. An admitted malicious or compromised tool can thereby convert untrusted data into recurring victim-billed processing without victim credentials or local runtime privilege. We call retained content persistent billable state and formalize the host's decision over whether and how it enters later billable context as the persistent billable-state boundary. We present the first systematic security study of this post-admission lifecycle. We derive six denial-of-wallet attack vectors and build DOW-BENCH, an end-to-end harness evaluated across six model families. Across 243 executions, usage telemetry shows that the maximum per-session cumulative input reaches 14,293x the session's first-call input. Controlled history-policy reruns isolate raw retention's contribution: retaining raw history increases mean effective session cost by 21.2-35.9%. Compression succeeds on 10/12 and 11/12 history-dependent tasks, versus 2/12 under deletion for each provider. To govern this boundary, we combine deterministic history transformation with four host-side invariants that bound prompt mass, context growth, recursive opportunity, and cumulative spend before reingestion. The kernel contains every recurring attack in the 123-evaluation replay corpus. Across 24 Mistral Small 4 workflows, a progress-authorized policy achieves 22/24 oracle-verified task successes with no pre-completion interruptions, versus 13/24 under a fixed cap. Only 71 of 3,830 scanned MCP server and transport repositories expose any code-visible safeguard proxy, and none cover all four safeguard families. These results establish persistent billable state as a first-class security object and pre-reingestion as its host-owned control point.

Multi-stage LLM-based cyber agents may complete attack workflows while remaining brittle, costly, or reliant on incorrect interpretations of execution evidence. Success rates alone obscure inefficiency, adaptation through retries, and recognition of success or failure. We present an end-to-end diagnostic study of an Autonomous Adversary system with orchestrator, executor, and validator LLMs in enterprise-like lateral-movement scenarios. Six frontier models are evaluated across two scenarios and three modes: expert-defined, self-scaffolded, and fully autonomous. We assess validator consistency and evidence grounding; introduce a subtask-conditioned, cost-aware score for abnormal token use, retries, and runtime; and use comparative LLM-as-a-Judge analysis to identify planning deficiencies, including tool misalignment, plan similarity, over-specification, inadequate probing, and weak recovery. Validators are generally relevant and evidence-grounded but often nonspecific and overly optimistic. Bottlenecks cluster in credential and lateral-movement tasks, spread with scenario complexity, and vary more under full autonomy. Reliable evaluation must assess outcomes, evidence interpretation, resource use, and adaptation after failure.

Agentic video-generation systems close a loop between a generator and a verifier: an LLM plans shots, calls a text-to-video model, and a multimodal judge decides whether the result satisfies the request. To diagnose where a long workflow fails, recent harnesses deliberately show the judge more than the video-the agent's execution trace, its plan, the narration it synthesized. We ask whether this auxiliary text moves the judge's verdict on purely visual requirements, holding the frames fixed. On a benchmark of 109 generated two-event clips with manual labels, in which the requested event is either visibly completed or visibly missing, a trace that reports a successful tool call makes three open-weight Qwen-VL judges (7B, 8B, 32B) accept $78$–$90\%$ of the failures, up from $7$–$19\%$ without text, and a contradicting trace makes them reject up to $100\%$ of correct clips; an instruction to “use only the frames” does not remove the effect. Frontier closed judges are essentially unmoved on the same clips, showing that the vulnerability is a property of the judge's learned trust in tool logs rather than of the task. Plan-derived text carries no clip-specific information, so it can only shift a judge's operating point, and in a repair loop that shift becomes a cap on the true pass rate that no repair policy can exceed; the cap matches simulation to two decimals. In the loop, contamination is exploited without any adversarial agent: an honest LLM planner that always regenerates ends with a judge pass rate of $1.00$ and a human-labelled pass rate of $0.28$, and a pipeline in which a cheap checker writes its verdict into the trace launders that checker's errors into a stronger final judge ($0.69$ false accepts).

LLMs increasingly operate through coding-agent harnesses that inspect repositories, invoke tools, and modify files. Substituting the model behind such an agent can therefore change security-relevant decisions, including whether it verifies changes or recovers safely from failures. Existing LLM fingerprints largely infer identity from direct text or token distributions. In coding agents, these signals are mediated by system instructions, controller logic, tools, and execution feedback, limiting their transfer. We present LIDAR (LLM Identification from Decisions and Actions at Runtime), an active black-box fingerprinting method for coding-agent execution. Three coding probe pairs expose post-edit verification, transient-failure recovery, and specification–test conflict resolution under controlled changes. LIDAR represents the resulting trajectories with complementary instance-level and distribution-level features and compares them with clean references using a lightweight probabilistic identifier. It requires no access to model weights, logits, or provider internals. Across 36 models from seven families and two agent harnesses, LIDAR achieves high Top-1 accuracy and MRR and outperforms four existing fingerprinting and API-auditing baselines. Ablations confirm that the two feature levels, all probe pairs, and their controlled variants contribute. These results show that agent execution behavior provides model-identity evidence beyond final outputs.

Humanoid control, leveraging human demonstrations, has achieved diverse, agile, and natural locomotion behaviors through reinforcement learning (RL). While this paradigm has yielded remarkable performance in physical humanoid control, how to eliminate specific motions from learned policies remains insufficiently explored. Addressing this issue is motivated by pressing safety and privacy concerns: the removal of malicious, poisoned, or suboptimal motions, as well as copyright-protected motions subject to the right to be forgotten under regulations such as the GDPR, is of critical importance. To this end, we propose {ForgetMimic}, the first motion-level unlearning method designed specifically for physical-world humanoid control. The core idea of ForgetMimic is as follows: given a policy $\pi_\theta$ trained on $N$ motions, our method degrades performance on a target subset of $K$ motions while preserving the effectiveness of the remaining $N-K$ motions. Furthermore, we identify and resolve two key training mechanisms in robot control that lead to unlearning failure. We conduct extensive experiments on the Unitree G1 and H2 humanoid robots across 12 motions, including Dance, Fight, Flip, and others. Experimental results demonstrate that ForgetMimic effectively eliminates memory of designated motions while maintaining the normal operation of all other motions.

Phishing emails that impersonate Nigerian fintech providers can combine deceptive sender addresses, lookalike links, and locally familiar language. This study presents a Gmail browser extension that integrates sender-domain and URL checks with a bidirectional long short-term memory (BiLSTM) classifier. The extension compares visible sender addresses and links with profiles for eight fintech platforms, obtains a phishing probability from a locally hosted Flask service, and displays a legitimate, warning, or phishing verdict when an email is opened. The BiLSTM classifier was evaluated on 8,943 test messages from a cleaned dataset of 59,622 phishing and legitimate emails. The test confusion matrix recorded 4,308 true negatives, no false positives, one false negative, and 4,634 true positives. These counts correspond to 99.99% accuracy, 100.00% precision, 99.98% recall, and 99.99% F1 score. Tokenized sequence analysis identified 5.79% overlap between the training and test sets, which may inflate performance estimates for independent messages. A Gmail demonstration showed the integrated extension producing user-visible verdicts, although the complete system was not evaluated on a labeled test set. The findings establish the feasibility of the implemented prototype while leaving its end-to-end detection performance and generalization to unseen attacks open for further evaluation.

We investigate $c$-interior pointwise maximal leakage (PML) as a tool for contraction analyses and disclosure control. Based on the strong adversarial threat models from maximal leakage, $c$-interior PML generalizes local differential privacy (LDP) to data-generating distributions with densities uniformly bounded away from zero by $c>0$. Viewing $c$-interior PML as an algebraic constraint on a kernel yields more flexible (and often tighter) contraction analyses than standard LDP. We provide tight bounds on the Dobrushin coefficient, and bound the contraction coefficient of the Hockeystick-divergence. We further derive strong data processing inequalities on $f$-divergences under $c$-interior PML constraints when the input distributions to the divergence are restricted to be in the $c$-interior. These results extend beyond the regime of pure LDP to cover a larger class of kernels, including, e.g., arbitrary stochastic matrices. We apply the results to minimax theory and provide asymptotically optimal strategies under $c$-interior PML constraints for binary hypothesis testing and mean estimation. The results show that disclosure control with PML allows analysts to reason about systems in a more differentiated manner: For example, it allows us to quantify the privacy leakage of deterministic systems, and can give precise adversarial guarantees with respect to arbitrary distributional assumptions. Interestingly, a recurring theme in the disclosure analyses is that if the privacy problem is relatively regular (if the density bound $c$ is large), private inference can be possible without incurring any additional cost in terms of sample complexity.