Papers

Recent work has applied differential privacy (DP) to adapt large language models (LLMs) for sensitive applications, offering theoretical guarantees. However, its practical effectiveness remains unclear, partly due to LLM pretraining, where overlaps and interdependencies with adaptation data can undermine privacy despite DP efforts. To analyze this issue in practice, we investigate privacy risks under DP adaptations in LLMs using state-of-the-art attacks such as robust membership inference and canary data extraction. We benchmark these risks by systematically varying the adaptation data distribution, from exact overlaps with pretraining data, through in-distribution (IID) cases, to entirely out-of-distribution (OOD) examples. Additionally, we evaluate how different adaptation methods and different privacy regimes impact the vulnerability. Our results show that distribution shifts strongly influence privacy vulnerability: the closer the adaptation data is to the pretraining distribution, the higher the practical privacy risk at the same theoretical guarantee, even without direct data overlap. We find that parameter-efficient fine-tuning methods, such as LoRA, achieve the highest empirical privacy protection for OOD data. Our benchmark identifies key factors for achieving practical privacy in DP LLM adaptation, providing actionable insights for deploying customized models in sensitive settings. Looking forward, we propose a structured framework for holistic privacy assessment beyond adaptation privacy, to identify and evaluate risks across the full pretrain-adapt pipeline of LLMs.

BCI-to-agent pipelines turn decoded neural activity into an authorization channel for tool-use agents, exposing a new attack surface we call brain-prompt injection: signal-side perturbations, context-only injections, and adaptive dual-decoder attacks can all change the routed action while EEG-side or text-side monitors remain blind. Route safety in this stack depends on what the audit log can observe, not on decoder accuracy or agreement alone. We define a Route-Safety Audit Contract: a minimal log schema, denominator hierarchy, and endpoint specification, and prove an audit-schema separation theorem together with a C3 attacked-dependence decomposition; clean agreement and marginal robustness do not identify the joint term that controls C3 routing. As a calibration layer on top of the contract, we apply split-conformal calibration to a non-oracle EEG confirmation channel and report the resulting false-accept frontier under an explicit threat-archetype matrix. We instantiate the contract on EEGMMI native left/right command-control over 5{,}400 events, harmless tool stubs, and seed/case denominators. Provenance blocks C2 routes ($0.000$); agreement-plus-provenance routes C3 flips ($1.000$); confirmation-plus-provenance routes them ($0.000$). The conformal frontier reaches FAR $0.000$ at clean utility $0.150$ for $\alpha=.005$ and FAR $0.119$ at clean utility $0.452$ for $\alpha=.10$ under acquisition isolation; an attacker-controllable confirmation channel breaks the bound to $\approx\!1$. Subject-cluster bootstrap confirms these intervals on $60$ subjects; cross-architecture (TinyEEGNet, EEGNetV4) and capacity-sweep results show within-regime saturation. Mediation and confirmation reduce risk; they are not intent certificates.

The convergence of the 2026 European Union Safe and Sustainable by Design (SSbD) framework, Corporate Sustainability Due Diligence Directive (CSDDD), and Carbon Border Adjustment Mechanism (CBAM) introduce a severe governance bottleneck for advanced semiconductor manufacturing facilities ("Smart Fabs"). Regulatory compliance demands have surpassed the capacity of manual corporate reporting, creating a direct conflict between multi-stakeholder transparency and corporate data privacy. This paper addresses this challenge by introducing a zero-trust socio-technical orchestration framework that operationalizes a six-layer SSbD reference architecture within trustworthy industrial data spaces. We propose a shift from reactive automation to autonomous governance through "Professional Proxies"-role-based agentic workflows executing within hardware-isolated trust zones. Structured as an interoperable network protocol stack, the framework coordinates an automated, five-step "relay race" between Facility, Process Engineering, and Finance proxy teams to align factory-floor yield models with macro-level sustainability mandates. By executing Virtual Metrology (VM) predictions and Federated Machine Learning (FML) inside hardware-rooted Trusted Execution Environments (TEEs), this architecture resolves the Data Sovereignty Paradox, demonstrating how fabs can export cryptographically signed compliance tokens via International Data Spaces (IDS) connectors without exposing proprietary process recipes. Ultimately, this framework provides technology managers with a verifiable, evidence-based pathway toward resilient, net-zero Industry 5.0 ecosystems.

We present a reproducible failure mode of safety training in RAG-based LLM recommendation, the Injection Paradox, in which prompt injections embedded in retrieved documents backfire against the attacker, suppressing the target brand below the injection-free baseline. In safety-trained Claude models, documents containing prompt injections suffer a sharp drop in recommendation rate, and this suppression propagates beyond the injected document to unmodified documents of the same brand. In Claude Opus 4.6, the target brand drops from a 54% baseline to zero top-2 recommendations across all 50 trials, even though only 1 of 4 brand documents in the corpus contains an injection. The directional pattern is reproduced in counterfactual experiments and across three brands. A contrasting result across the GPT models tested, where the same injection instead increases recommendations, suggests model-family differences in how injection-like context affects recommendation behavior. These findings raise the technical possibility of a reverse-attack scenario in which an adversary embeds injections in a competitor's documents to suppress the competitor's brand via safety-sensitive model behavior. Code, prompts, privacy-preserving per-trial outcome records, and aggregate results are released.

EEG foundation-model releases are usually audited one endpoint at a time: raw-reconstruction, membership inference, identity linkage, or DP-SGD on the downstream head. We audit the same released embeddings under all four endpoints jointly, on BIOT, LaBraM, and EEGPT, and show that each single-endpoint audit clears releases that still leak spectral attributes. The decisive evidence is a cross-encoder transfer audit: a single ridge attribute decoder learned from one frozen encoder transfers, via a fitted linear bridge, to held-out-subject test splits of every other encoder, with subject-disjoint matched-control 95% CI lower bound at least 0.081 across all six BIOT/LaBraM/EEGPT directions. We prove a sufficient condition: two encoders sharing a nontrivial attribute-coordinate projector overlap beta admit a chained ridge bridge attacker with centered-gain lower bound sqrt(beta/(1+tau^2)) - eps_br - rho_0, and back-solve beta in [0.008, 0.198]. To turn the joint audit into a deployment-readable decision rule we introduce an audit-endpoint disagreement score (AEDS), prove sufficient conditions for its positivity, and bootstrap-calibrate it per cell; AEDS is positive in all eight matched-CI cells (BIOT/LaBraM/EEGPT on EEGMMI; LaBraM on Sleep-EDF, 54-channel LIMO, CHB-MIT pediatric scalp EEG) with p<0.001, while a head-level Carlini LiRA membership audit reaches AUC only 0.50-0.70. Standard defenses fail under audit: a Wiener-style noise-aware adaptive attacker, the LiRA audit, and DP-SGD at every utility-preserving epsilon in {4,8} leave the attribute channel essentially unchanged. The contribution is an audit framework that turns scattered single-endpoint defenses into a joint release decision, supported by a cross-encoder bridge theorem and adaptive-attacker, LiRA, and DP-SGD baselines; the audit licenses release-blocking, not raw-waveform exfiltration or held-out-subject identity recovery.

EnclaveScale is a distributed, hardware-assisted telemetry architecture providing post-extraction attestation, enabling operators to collaboratively model high-resolution generative AI power transients. Existing cryptographic techniques scale poorly for 10-Hz streaming or fail to authenticate origins, permitting malicious hosts to spoof sensor inputs. We implement and evaluate a post-extraction pipeline utilizing DCAP attestation, differential privacy noise injection, and Byzantine rejection across 32 GCP Confidential VMs, achieving 0% post-extraction attack success rate. This edge-DP approach distils continuous GPU transients into discrete Markov-chain transition matrices, guaranteeing event-level differential privacy. To mitigate pre-ingestion vulnerabilities, we propose an SPDM-authenticated first-mile layer. While current platforms lack attested I/O, emerging hardware architectures integrate PCIe IDE and TDISP to natively prevent host-level synthesis, securing the end-to-end provenance boundary. A Global Aggregation Enclave verifies these cryptographic proofs prior to capacity-weighted aggregation. Evaluation demonstrates a steady-state throughput of $131{,}406$ samples/s per enclave, amortising attestation overhead to $0.23\,\mu$s/sample. On empirical NVML-sampled H100, A100, and L4 traces, EnclaveScale achieves a dynamic orchestration margin error of $1.3$\,MW compared to $0.1$\,MW for an honest-aggregator central-DP baseline. EnclaveScale establishes a secure foundation for dynamic multi-tenant power orchestration, obfuscating sub-second anomalies locally and protecting macro-workload confidentiality via spatial dilution during global aggregation.

The prosperity of text-to-image (T2I) models has fostered a vibrant share-and-play ecosystem centered on Low-Rank Adaptation (LoRA) plugins, which allow users to customize and share model capabilities with ease. This democratization, however, comes with a hidden but severe security risk. Malicious users could share and distribute seemingly benign LoRA plugins that contain hidden functionalities to poison the model-sharing market, like Civitai or Liblib, severely undermining the user trust that underpins this collaborative ecosystem and threatening the safety of countless downstream applications. Despite these risks, plugin poisoning in the real-world T2I ecosystem remains underexplored. This paper introduces PoisonLoRA, the first systematic study of LoRA plugin supply-chain risks that exploits the trust and characteristics within the T2I ecosystem. We identify two primary attack instances: (1) Concept Hijacking, where a hijacked LoRA could generate images to influence public opinion and spread propaganda, and (2) Task Injection, where a LoRA is injected to produce harmful content (e.g., NSFW images) only activated by a secret key. Critically, the malicious payload persists with virus-like propagation. Such propagations weaponize the very act of creative collaboration (e.g., LoRA merging) to spread its contagion, turning every remix into a new carrier. Extensive experiments validate that PoisonLoRA is both effective and stealthy. Specifically, we achieve approximately 100% attack success rates (ASR) on both Civitai and Liblib on 6 datasets across 4 scenarios, without being detected by the platforms. The poisoned LoRA demonstrates extreme robustness, with nearly 100% ASR even transferred to different base models and remixed more than 5 times.

Large language models fine-tuned on instruction-code pairs may memorize and subsequently leak sensitive training data. Existing differentially private (DP) code generation methods primarily protect code snippets while assuming prompts are public, which fails in realistic scenarios where prompts may also contain sensitive information. When prompts cannot be explicitly learned or used during generation, code synthesis suffers from severe utility degradation as well as reduced diversity and fidelity. To address these challenges, we propose PrivCode-Plus, the first work to explore DP code generation where both prompts and code snippets are considered sensitive in LLM fine-tuning. PrivCode-Plus introduces a two-stage DP framework with a Privacy-Free Latent Conditioning module, enabling effective DP fine-tuning and data synthesis without direct access to sensitive prompts or code. Extensive experiments show that PrivCode-Plus achieves substantially higher utility than baselines, remains competitive with the method with relaxing privacy assumptions, and provides stronger privacy guarantees.

We demonstrate that widely deployed Large Language Model (LLM) inference stacks harbor a steganographic channel that requires no modification to model weights, sampling code, or output distributions. The channel exploits a structural property of deterministic decoding: pseudo-random number generators (PRNGs) used in inverse-transform sampling produce a seed-dependent sequence of token-level probability intervals that can be reconstructed from the generated text alone. A sender encodes a secret message in the PRNG seed before generation; a receiver reconstructs the intervals and recovers the seed, and thus the hidden payload, by exhaustive search over the seed space. We formalize two operational modes. In the known-prompt setting, sender and receiver share the prompt, enabling exact interval reconstruction and perfect seed recovery via forced alignment. In the unknown-prompt setting, only the generated text is available; approximate interval reconstruction combined with a maximum-hit-count scoring strategy still permits reliable recovery from sufficiently long outputs. Extensive experiments across six model families and five heterogeneous text domains show that, in the known-prompt setting, full 32-bit seed recovery from the complete 2^32 candidate space achieves up to 100% accuracy, depending on model and text domain, within 300 tokens and under 35 seconds on a single GPU. In the unknown-prompt setting, recovery reaches near-perfect accuracy at 600-800 tokens in about 12 seconds. We further analyze the influence of prompting strategies, tokenization ambiguities, and sampling hyperparameters on channel reliability. Moreover, we discuss several applications of our results: First, it allows for the steganographic transmission of 32 bits, but also shows that ignorance of the prompt is not a valid security assumption.

Privacy risks in text-only Large Language Models (LLMs) are well studied, particularly their tendency to memorize and leak sensitive information. However, Multi-modal Large Language Models (MLLMs), which process both text and images, introduce unique privacy challenges that remain underexplored. Compared to text-only models, MLLMs can extract and expose sensitive information embedded in images, posing new privacy risks. We reveal that some MLLMs are susceptible to privacy breaches, leaking sensitive data embedded in images or stored in memory. Specifically, in this paper, we (1) introduce MM-Privacy, a comprehensive dataset designed to assess privacy risks across various multi-modal tasks and scenarios, where we define Disclosure Risks and Retention Risks. (2) systematically evaluate different MLLMs using MM-Privacy and demonstrate how models leak sensitive data across various tasks, and (3) provide additional insights into the role of task inconsistency in privacy risks, emphasizing the urgent need for mitigation strategies. Our findings highlight privacy concerns in MLLMs, underscoring the necessity of safeguards to prevent data exposure. Our dataset and code can be found here.

Tool-using LLM agents interact with the world through actions that persist state in artifacts (e.g., workspace files or logs). Consequently, jailbreak defenses must reason about cross-step composition rather than isolated text. Yet most existing attacks and defenses, including “multi-turn” jailbreaks such as Crescendo and Tree of Attacks,still assume a single contiguous conversation visible to the defender. This assumption breaks down in real agent pipelines, where enforcement is fragmented across tools, modules, and time, and where artifact provenance is often not tracked. We operationalize a deployment failure mode for tool-using LLM agents, the provenance gap, and study reproducible triggers for it: Context-Fractured Decomposition (CFD), a family of cross-context multi-step jailbreaks that preserve benign-looking intermediate artifacts from an early interaction and elicit harmful behavior much later, potentially in a different agent instance or workflow stage, via individually innocuous tool actions whose risk emerges only under delayed artifact-mediated composition. We instrument the failure mode with trace-level diagnostics and outline a verifiable mitigation direction (provenance lineage tagging). Across agent-system jailbreak benchmarks, CFD improves success rates by up to 28.3 percentage points over state-of-the-art baselines, even against strong single-turn judges. Disclaimer: This paper contains examples of harmful or offensive language.

Modern enterprises face an accelerating onslaught of API-targeted threats amid a rapidly expanding attack surface. Record volumes of software vulnerabilities continue to accelerate dramatically, with 28,818 CVEs disclosed in 2023 (a 38% jump from 2022) and 40,009 CVEs in 2024 (another 38% increase), while the average time-to-exploit (TTE) of new flaws shrank to mere days (approximately 5 days in 2023, down from 32 days in 2021). At the same time, API usage dominates web traffic and has become a primary vector for breaches - 99% of organizations experienced API security incidents in the last year, with 22% suffering actual data breaches via APIs (based on industry vendor research). This paper proposes a comprehensive "security-first" framework for API pipeline development, leveraging Zero-Trust Architecture principles within DevSecOps practices to counter these trends. We introduce a five-pillar approach encompassing Governance & Planning, Secure Design, Continuous Testing, Pipeline Controls, and Runtime Protection, aligned with industry standards (OWASP API Security Top 10 2023, NIST Secure Software Development Framework) and recent cybersecurity advisories. The results show significant improvements in vulnerability mitigation and breach prevention (e.g., 30% reduction in security incidents and 40% fewer post-release vulnerabilities in representative case studies), highlighting the positive impact of proactive security integration. The paper concludes with a discussion on implementation challenges, the evolving threat landscape, and recommendations for organizations to adopt a security-first pipeline with Zero-Trust to fortify API development against current and future threats.

Retrieval-augmented generation (RAG) systems often serialize user queries, retrieved documents, metadata, system labels, and task instructions into one natural-language prompt. We study a source-authority boundary failure in this design: attacker-authored retrieved text can impersonate metadata, provenance, authority, or disclosure-policy signals that appear control-relevant to the model. We call this pattern Document-Authored Control-Signal Impersonation (DACSI). DACSI is a non-imperative, metadata-like payload subclass within indirect prompt injection. Its central lesson is simple: document-authored labels are data, not policy. Command-style injection asks the model to ignore, override, or violate policy; DACSI asks whether untrusted document text can be misattributed as an authorized control signal when RAG prompt rendering collapses trusted and untrusted text into the same natural-language channel. We evaluate DACSI across six model settings, prompt-pressure levels, injection baselines, signal taxonomies, RAG-mediated pipelines, system-control probes, a source-authority attribution probe, and synthetic canary formats. We interpret the evidence by model regime rather than as six equal replications: DeepSeek V4 Pro and Qwen3.5-397B provide the cleanest positive lift, DeepSeek V4 Flash is a high-susceptibility setting, GPT-5.5 and Gemini 3.1 Pro Low are strong-boundary probes with selected residual risks, and GLM-4.7 is a saturated leakage boundary case. Across these regimes, DACSI warrants separate evaluation because it uses a command-free metadata/provenance/policy surface, follows a RAG-specific source-authority path, and responds to source/channel separation. The source-authority probe is behavioral attribution evidence, not proof of an internal mechanism.

Agent benchmarks score submissions with outcome verifiers that are typically hand-written and brittle, leaving them open to reward hacking. We audit 1,968 tasks across five terminal-agent benchmarks and find 323 (16%) hackable by frontier models given only the task description. This corrupts both leaderboard rankings and RL training signal, yet the standard response is manual and reactive. We introduce the hacker-fixer loop, a method for building exploit-resistant verifiers without per-task manual patching. The loop alternates three LLM agents: a hacker tries to pass the verifier without solving the task, a fixer patches the verifier to reject each discovered exploit, and a solver confirms the patched verifier still admits legitimate solutions. The loop iterates: each patch reshapes what the verifier rewards, surfacing the next exploit. We further add verifier access, and let patches transfer across tasks, to broaden the exploits the loop discovers. On KernelBench, the loop drives the attack success rate from 62% to 0% on a held-out corpus of publicly reported exploits. We also find that weaker agents in the loop can defend against much stronger hackers: Gemini 3 Flash's loop drives the stronger Gemini 3.1 Pro and Claude Opus 4.7's attack success rate from 76% and 61% to 0% on KernelBench, and Gemini 3.1 Pro's from 39% to 17% on Terminal Bench across 77 tasks. We release Terminal Wrench (323 hackable environments, 3,632 hack trajectories) as a snapshot of the current attack surface, our patched verifiers, the exploits the loop discovered, and our implementation as a basis for future work.

As LLM agents begin to take real, irreversible actions (shell commands, file edits, deploys), the standard safety pattern is a human-in-the-loop approval gate: risky actions pause and wait for a person. We argue the gate is the easy part; the hard part is the judgment - which actions to stop - which the field evaluates against two false assumptions: that there is a ground-truth notion of "risky," and that the human reviewer is a perfect, infinitely-available oracle. On a hand-labeled set of 125 adversarially-weighted agent actions we show that (i) reviewers only moderately agree on what is risky (Fleiss' kappa = 0.52), so there is no single correct label; (ii) framing the guard as selective classification under asymmetric cost makes its operating limits measurable, and on hard inputs the guard cannot safely auto-decide; and (iii) when the reviewer is modeled as endogenous (fatiguing as escalation load grows), realized safety becomes an inverted-U in the escalation rate: more human oversight can make a system less safe, and the safety-optimal guard escalates below full escalation - a setting a load-aware policy also uses to resist a flooding attack that slips a malicious action past a fatigued reviewer. Agent oversight, framed this way, is not only a classification problem but a resource-allocation one: human attention is finite, and the guard's escalation policy spends it. We claim none of these mechanisms as novel - fatigue-aware learning-to-defer (FALCON), cost-sensitive deferral under workload constraints (DeCCaF), trajectory-level guarding, and reviewer-fatigue/flooding attacks are all prior art we cite. Our contribution is an open-source agent-oversight system that operationalizes and measures them in the LLM-agent action-gating setting, turning "is my guard good?" from a guess into a curve. The inverted-U and the flooding attack are modeling results that motivate a human study.

A small transformer encoder is trained to map Terminal-Wrench trajectories onto a unit sphere where embedding distance approximates the $L_1$ distance between reward and metadata signals. A linear probe on top of that embedding detects reward hacking on the cleaned test split with AUC $0.9467$ and TPR@5%FPR $0.8296$, matching the TW sanitized LLM-as-judge AUC ($0.9510$ on the cleaned split) and exceeding its TPR@5%FPR ($0.7130$ vs $0.8296$) on the same information condition, at roughly four orders of magnitude lower per-trajectory cost. The encoder is not a pure behavior reader: stripping natural-language reasoning from its input at probe time drops AUC to $0.6213$.

Internet censorship affects over four billion people, and deployed circumvention systems share a common weakness: their endpoints are fixed and discoverable, so a patient censor can enumerate and block them. Moving-target circumvention systems instead rotate endpoints across commercial cloud address space faster than censors can react, but the field lacks a theory of when rotation works, leaving rotation intervals and pool sizes to intuition. We give the first formal account of moving-target censorship resistance by modeling the censor-defender interaction as a continuous-time timing game over a combinatorial address-domain space, generalizing FlipIt to a collateral-bounded adversary. We prove a sustainability frontier separating configurations a censor can defeat from those it cannot, and show that under the Great Firewall's 2024 shift to blocking QUIC and TLS by domain, raw rotation speed is not the binding constraint. Instead, availability is governed by the domain burn rate, $\beta=\lambda_{\mathrm{disc}}/\lambda_{\mathrm{intro}}$, the ratio between how quickly the censor blocks defender domains and how quickly the defender introduces fresh ones. We derive a closed-form availability law, prove that address rotation alone cannot sustain high availability when $\beta>1$ regardless of endpoint rotation speed, and characterize the frontier $\beta^\star$. We validate the analysis with an open, model-level censor-defender simulator requiring no privileged access or cloud deployment. The simulator reproduces the predicted phase transition at $\beta^\star$ under adversary profiles representative of the GFW, Russia's TSPU, and Iran, and shows robustness to state-dependent discovery and bursty, provider-correlated burns. The result replaces the heuristic of “rotate faster” with a precise operating condition: keeping the domain economy ahead of the censor.

Abliteration, the removal of refusal capabilities from large language models by projecting weight matrices orthogonal to an extracted refusal direction, has emerged as a prominent safety concern through its ability to bypass post-training alignment using only a small set of contrastive prompts. We find that existing defenses commonly overlook the cause of abliteration; that is, how easily the refusal direction can be extracted. To hinder this process, we introduce a weight-editing method that obscures the refusal signal by applying rank-$k$ updates to residual stream writer matrices while replacing refusal-inducing activations with random aliases and correcting downstream reader matrices to preserve the model's original behavior. On Llama-3-8B, AMRA improves post-abliteration refusal scores by $2.16$ points over the undefended baseline with less than $0.5$ percentage points of MMLU degradation. On Gemma-2-9B, it improves the post-abliteration refusal by $14.70$ points over the baseline while keeping harmful output rates similar to the baseline, albeit at a greater utility cost.

AI-powered agents are increasingly embedded in continuous integration and continuous delivery/deployment (CI/CD) pipelines to autonomously review pull requests (PRs), triage issues, and maintain codebases. These agents ingest untrusted content while operating with elevated repository permissions, making them a natural target for prompt injection attacks with supply chain consequences. We present GitInject, an open-source framework for evaluating prompt injection vulnerabilities in real, live GitHub workflows, a widely deployed instance of CI/CD pipelines. Unlike prior agent security benchmarks that simulate tool calls, GitInject provisions ephemeral repositories and triggers actual workflow runs, so that sandbox constraints, credential handling, and permission boundaries behave exactly as in production. Using GitInject, we study workflow configurations across four AI providers and document eleven named attacks spanning config-file injection, credential exfiltration, judgment manipulation, and availability. We find that all tested providers are susceptible to at least one attack class in their default configuration, and that the most critical vulnerabilities are structural: they arise from how CI/CD infrastructure handles credentials and configuration files, not from any specific model's behavior. For each confirmed attack class, we identify the minimum-cost workflow-level countermeasure and analyze its coverage and limitations. GitInject is released publicly to facilitate further research in this direction.

Feature selection is critical for network intrusion detection systems (NIDS) operating under high-dimensional, highly imbalanced traffic, as found in operational and defense networks. Traditional filter methods rank features using global statistics computed symmetrically across classes and thus fail to capture the asymmetry of intrusion detection, where attacks are best characterized as deviations from dominant benign traffic. We propose benign-anchored Classwise Mean Deviation (nCMD), a lightweight and interpretable method that scores feature relevance based on the deviation of attack-class distributions from the benign-class mean, rather than a globally biased reference. This approach aligns feature selection with the operational semantics of NIDS at no additional computational cost. Across four benchmark datasets (CICIDS2017, CICDDoS2019, NSL-KDD, and UNSW-NB15), multiple feature budgets, and three downstream classifiers, nCMD matches or exceeds classical filter baselines in macro-averaged F1-score. It achieves the best result on three of the four datasets and under every classifier, with the strongest improvements observed under tight feature budgets and severe class imbalance. These results support benign-anchored ranking as a scalable and interpretable preprocessing component for resource-constrained NIDS.

Autonomous agents negotiate, purchase, deploy code, and move funds, but no neutral mechanism determines whether they met their delegated obligation, who is responsible when they did not, or which settlement action follows. This is the agentic clearing problem. Tool protocols (MCP), inter-agent communication (A2A), payment rails (x402), mandate and network agent protocols (AP2, Visa, Mastercard), and settlement-risk standards each assume that determination and none produce it. Clearing is the missing primitive. Payment is not clearing. Authorization is not clearing. LLM-as-judge evaluation is not clearing. Settlement-risk escrow is not clearing: it consumes clearing decisions. RAILS (Real-Time Agent Integrity & Ledger Settlement) is the integrity and clearing layer for agentic commerce, spanning a per-output reliability score, a published reliability record, and a clearing function that consumes them. The clearing protocol at its core closes that gap. Seven primitives (Obligation Object, Evidence Envelope, Verification Mesh, Clearing Decision, Settlement Instruction, Clearing Passport, Finality Rules), bound by a formal model of admissibility-graded verification, together yield a soundness property: no financially material settlement is supported by evidence below the obligation's admissibility floor. The property is falsifiable against the spec. We are not aware of a prior agent-commerce verification mechanism that states a property of this kind. The approaches nearest to it emit a pass, a delivery guarantee, a bare score, or an equilibrium. This paper specifies that clearing protocol.

The aim of steganography is to hide secret information inside ordinary media so that the existence of communication is hidden rather than encrypted. In audiovisual context, the availability of audio and video streams creates an opportunity to split a payload across these two modes thus, reducing the embedding burden on any single carrier. This paper evaluates whether such split-payload audiovisual steganography can help evade unimodal and multimodal steganalysis under synchronized and asynchronous embedding settings. We create audiovisual samples where the hidden message is divided between the audio and video tracks, and then test how well different detectors can identify them. The single mode detectors performs close to random guessing, thus showing the benefit of this hiding mechanism, while the multimodal model initially appears more effective. However, further checks show that this improvement mostly comes from the video stream, not from a true combined audio-video signal. Overall, the results suggest that splitting the payload across modalities can make detection harder, but multimodal detectors must be evaluated carefully to ensure they are learning the intended signal.

Structured Cyber Threat Intelligence (CTI) increasingly supports adversary emulation, detection evaluation, and cyber range design, yet each workflow still requires a target System Under Test (SUT) whose environment is not fully described by public CTI. We define the environment semantics gap as a measurable property of structured CTI: the SUT information required for replay-ready instantiation that cannot be derived solely from structured fields. We present AutoSUT, a pipeline that locates where corpus-supported narrowing ends and analyst specification begins. Across ATT&CK Enterprise, Mobile, and ICS STIX bundles, with CAPEC and FiGHT as contrast datasets, we measure platform coverage, software specificity, vulnerability evidence, and deployment compatibility. Platform tags are near-universal, but 97.6% of Enterprise software objects lack version indicators and CPE identifiers. Campaign-level CVE evidence covers only 9.6% of campaigns, even after free-text enrichment, and only 19 of 691 techniques (2.7%) are container-feasible under conservative backend-family assignment. Profile confusion among intrusion sets drops from 1.3% for one linked software item to 0% for two linked software items, indicating that software-evidence density, not CVE enrichment, drives actor-specific SUT screening. Finally, we constructively demonstrate environment non-uniqueness: holding every corpus-supported element fixed and varying only the analyst-authored region yields multiple distinct, campaign-compatible SUTs, including an executable witness running CVE-2021-41773 and coincident witnesses in which structurally different service realizations execute the same attack. Structured CTI, therefore, constrains but does not uniquely determine the executable environment. Replay-ready emulation should accordingly declare which environment commitments the corpus supports and which remain analyst-authored.

The additive noise mechanism is a foundational tool for differential privacy (DP) of $T$-dimensional real-valued vector queries. The Gaussian mechanism, utilizing Gaussian noise, is the mostly widely used such mechanism, due to its simplicity and strong privacy guarantees. In this work, we provide justification for this choice, showing that as the dimension $T\to\infty$, no additive-noise mechanism can asymptotically improve on the Gaussian mechanism's privacy–utility tradeoff for the strong privacy settings typically used.We also develop a new family of Spherical Generalized Gamma DP mechanisms, which contains both the Gaussian mechanism and the recently studied $\ell_2$ mechanism (Joseph et al., ICML 2025). We identify members of this family that outperform both the Gaussian and $\ell_2$ mechanisms in certain low-dimensional settings, and show tight composition of all mechanisms in this family, answering an open question of Joseph et al.~regarding the $\ell_2$ mechanism.

Age verification is rapidly emerging as a central regulatory instrument for protecting minors online, with several jurisdictions mandating its deployment for access to adult and pornographic content. This regulatory direction raises significant privacy concerns, as it risks binding sensitive content access to identity-related attributes. It also introduces security risks, since age-verification mechanisms are often outsourced to third-party providers with limited transparency into the robustness of their verification processes. In this work, we conduct, to the best of our knowledge, the first exploratory security assessment of regulation-mandated age-verification mechanisms deployed by adult websites. Rather than treating age verification as a purely regulatory question, we empirically examine whether current deployments provide security guarantees commensurate with the privacy risks of relying on sensitive identity-related data. Our methodology combines ecosystem mapping, adversary modeling, and empirical testing across four countries, covering document-based verification, biometric age estimation, indirect signals, and website-workflow integration. Our results reveal systemic weaknesses across mechanisms and integrations under realistic threat assumptions, including failures against low-cost, widely accessible attacks. Finally, we derive concrete guidelines and design directions for mitigating the security and privacy risks exposed by current age-verification deployments.

Motivated by polynomial identity testing with exponentials (Li and Wu, ITCS'26), we study uncertainty principles for the number-theoretic transform (NTT). We show that the NTT satisfies strong sparsity tradeoffs: For every fixed prime $q$ and for all but finitely many primes $p \equiv 1 \pmod q$ every nonzero $f\in \mathbb F_p^{\mathbb Z_q}$ and its number-theoretic transform $\hat f$ satisfy \[ |\mathrm{Supp}(f)| + |\mathrm{Supp}(\hat f)| \ge q+1. \] Thus, a $k$-sparse function has transform support at least $q-k+1$. As our main technical contribution, we prove a probabilistic version of the above uncertainty principle, averaged over primes $p$, in the regime $p=q^{O(1)}$. As an application, we obtain a black-box identity test for $k$-sparse exponential polynomials of degree at most $d$ with vanishing soundness error, for $q$ moderately larger than $k$.

Data agents integrate LLM-driven reasoning with relational data access, executable analytical tools, and multi-step workflow orchestration, making them increasingly central to enterprise analytics. This integration introduces new security vulnerabilities across data resources, database execution, and agent reasoning, recombining concerns from database security and general-purpose LLM-agent security into failure modes that neither line of work captures on its own. To address this gap, we present a systematic security study of data agents. Our contributions are threefold. First, we develop a layered vulnerability framework that identifies eight data agent-specific risks across interpretation, execution, and policy layers. Second, we introduce an attack taxonomy organized by adversary goal, tactic, and technique, covering three goals, seven tactics, and fourteen techniques, and pair it with an LLM-driven payload generation pipeline grounded in real database schemas. Third, we evaluate these attacks on six systems, including four open-source data agents and two production cloud analytics services. Our experiments reveal substantial security vulnerabilities across current systems and yield four key takeaways.

Forensic analysis of web server logs demands both accurate detection and human-readable explanations that can satisfy legal requirements. We present CEF-Log, a context-enhanced few-shot chain-of-thought prompting strategy for Large Language Models that addresses this dual requirement. CEF-Log embeds expert investigative methodology through a structured five-step reasoning template, enabling the model to learn how to analyze logs rather than what patterns to memorize. Experimental evaluation demonstrates that CEF-Log achieves an F1-score of 0.99 on the CSIC 2010 dataset using only four examples while providing a $10\times$ improvement in sample efficiency compared to other prompting-based methods. We also introduce ForenWebLog, a new dataset that incorporates real-world attacks and multi-step attack sequences for comprehensive evaluation. Qualitative analysis confirms that CEF-Log generates traceable, accurate explanations suitable for forensic documentation, addressing the critical "black-box" limitation of traditional machine learning approaches.

Privacy-preserving Personalized Federated Learning (PFL) enables clients to collaboratively train personalized models without exposing raw data, but exchanged model updates remain vulnerable to inference attacks from honest-but-curious servers. Homomorphic Encryption (HE) addresses this by allowing server-side aggregation directly on encrypted updates, with the CKKS scheme being particularly suitable due to its native support for approximate floating-point arithmetic. However, no prior work has examined how to configure CKKS for PFL deployments, leaving practitioners without principled guidance on parameter selection that directly affects privacy, precision, and computational cost. This paper presents pFedCKKS, a generic framework integrating CKKS into PFL, and provides the first systematic parameter selection guide for practitioners. We derive the full CKKS parameter constraints under 128-bit security for the PFL setting, showing the selection problem reduces to choosing just two values: the inner and outer ciphertext prime. Implemented using the Flower framework and TenSEAL library, pFedCKKS is evaluated on the FEMNIST, CelebA and Sentiment140 datasets with FedFinetune, Ditto and FedPer which represents PFL algorithms. Experimental results reveal an empirical trade-off between precision and computational/communication costs. This allows us to draw a concrete guideline for selecting proper CKKS parameters that balance efficiency and accuracy in real-world deployments of pFedCKKS.

Mobile phone overuse and attention fragmentation have become pressing societal and public health concerns. Cyberpsychology research highlights addictive engagement loops driven by intermittent rewards, persuasive design, and habit formation. In this article, we use current evidence on mobile-phone addiction and propose "Digital White Spaces" (DWS), a socio-technical framework that combines privacy-preserving monitoring, AI-driven detection of addictive loops, device-mode interventions, and physical signal-limited zones to minimize digital stimulation and internet addiction.