Papers

Unsolicited automated phone calls (robocalls) are a serious threat: in the US alone, these calls resulted in reported losses of 1.1$ billion during 2025. Phishing and spoofing consistently rank among the most reported crimes within the FBI's Internet Crime Complaint Center, with phone call scams having the highest reported median loss. Combating robocalls is difficult due to many legal and practical constraints: robocalls often encompass multiple legal jurisdictions of different countries/states, the large volume of robocalls, their multilingual nature, the lack of publicly available data, privacy concerns with obtaining data, etc. We present a study of international robocalls, aggregating robocall reports from countries across all inhabited continents and contribute by providing new findings on international robocalls from 65 different countries. We also present the first publicly available multimodal and international robocall dataset: 8.7 million call detail records, 839 robocall transcripts from 28 identified robocall campaign clusters, and 677 robocall recordings. We describe our methodology for collecting robocall data over a 9-month period and provide a detailed analysis comparing robocalls in the US with those in other countries. Our analysis covers several aspects, including uncovering calling patterns, identifying co-targeting attacks, discovering common robocall campaigns, extracting callback numbers, analyzing linguistic differences among robocalls in the same language but different regions, and other insights. Our results indicate that although robocalls are an international problem, the severity of the threat is significantly higher in the US than in other countries. We provide steps for future research and suggest remedies to reduce the effectiveness of robocalls based on our analysis.

As ML-KEM is adopted as a post-quantum cryptographic standard, resilience against physical side-channel attacks has become essential. Among the constituent steps, the decapsulation Fujisaki-Okamoto (FO) verification is particularly vulnerable to side-channel power and electromagnetic (EM) analysis. In this work, we focus on common FPGA-based implementations and examine their side-channel vulnerabilities, and compare them with those of microcontroller implementations. Three verification implementations, unprotected, hash-based (first-order), and higher-order masked, are evaluated for side-channel security on both a microcontroller and an FPGA. While FPGAs offer higher speed and parallelism, they often exhibit stronger side-channel leakage, especially in high bandwidth configurations. The higher-order masked designs still leak information about the underlying data due to hardware-level effects and data-dependent processing. Our experiments show that their parallelized processing on FPGAs introduces sufficient first-order leakage for full secret-key recovery. These results underscore the persistent challenge of securing PQC algorithms in performance-constrained and parallelized hardware environments.

Large language models are no longer only text generators. They are increasingly embedded in retrieval pipelines, enterprise assistants, coding environments, robotic systems, security-operation workflows, and autonomous agents that can read private data, call tools, write files, execute code, and act across organizational boundaries. This shift changes the security problem: risks do not arise from the model weights alone, but from the full lifecycle and application stack through which data, prompts, model outputs, tools, memories, and user authority interact. This paper systematizes the literature on vulnerabilities in large language model systems through a lifecycle and application-stack lens. We organize attacks across eight stages: data collection, pretraining, post-training alignment, model packaging and supply chain, retrieval and memory, prompting and inference, tool/agent execution, and deployment/maintenance. For each stage, we analyze attacker capabilities, affected security objectives, representative attacks, practical risks, evaluation practices, and defenses. We further map LLM-specific vulnerabilities to confidentiality, integrity, availability, safety, privacy, fairness, accountability, and agency-control objectives. Unlike taxonomies that list isolated attack names, the proposed systematization emphasizes where trust boundaries fail, how untrusted data becomes executable instruction, how delegated authority amplifies model errors, and why point defenses rarely compose. We close with a research agenda for secure LLM systems, including compositional security, provenance-aware retrieval, tool-call containment, long-horizon agent evaluation, privacy-preserving adaptation, realistic red teaming, and deployment-grade incident response.

Network intrusion detection systems (NIDS) are crucial in cybersecurity infrastructure, needing advanced techniques to detect hostile activity in network traffic. This research introduces a hybrid approach that combines Topological Data Analysis (TDA) with Long Short-Term Memory (LSTM) networks to improve anomaly detection in network security. Our multi-layered design combines TDA's persistent homology with LSTM networks to capture topological characteristics of network traffic patterns and simulate temporal sequences. We assessed our methodology using the CIC-IDS2017 dataset, which includes over 2.8 million labelled flows, 77 network variables, and 14 attack categories that reflect modern threat landscapes such as DDoS, brute force, web attacks, penetration, and botnet activities. Integrating Betti curves and persistence diagrams with deep learning architectures enhances feature extraction performance. Our hybrid TDA+LSTM model has an AUC of 1.000 and F1-score of 1.000, with 5-fold cross-validation producing a mean AUC of 1.000 $\pm$ 0.000 and mean F1 of 0.999 $\pm$ 0.001. An ablation research demonstrates the complimentary contributions of topological (F1=0.990) and temporal characteristics (F1=1.000). Comparative research shows that the suggested strategy beats TDA+Random Forest (F1=0.994) and Isolation Forest (F1=0.835) baselines in several attack categories.

This work presents Dual-Embedding Watermarking (DEW), a semantic watermarking scheme for large language models (LLMs) that leverages contextual and token-level embeddings to enhance robustness against paraphrasing and translation. DEW utilizes a signal-processing methodology, applying algebraic vector-space operations to token and context embeddings to derive a watermark signal that degrades gracefully under semantic shifts. The method obfuscates the watermark by projecting embedding vectors through pseudo-random matrices seeded with a secret key. Experimental results show that dual-embedding watermarking can offer state-of-the-art robustness, particularly against translation, while incurring relatively low computational overhead compared with other semantic schemes. At lower watermark strength, DEW also maintains competitive text quality, suggesting that dual-embedding signals provide a promising substrate for robust semantic watermarking.

Digital signature schemes are an important cryptographic tool to ensure data authenticity and integrity in many applications that must be resilient to attacks, including those facilitated by quantum computers. We consider the two digital signature schemes based on error-correcting codes that are second-round candidates in NIST's call for Additional Signature Schemes, which is part of the Post-Quantum Cryptography Standardization Process. Specifically, we provide an overview of the Codes and Restricted Objects Signature Scheme (CROSS) and the Linear Equivalence Signature Scheme (LESS). We describe their underlying problems of syndrome decoding from restricted errors and code equivalence. We review sigma protocols and how they can be transformed into digital signature schemes via the Fiat-Shamir transform. Finally, we explain how this procedure yields code-based digital signatures believed to be post-quantum secure.

The Internet of Things (IoT) is rapidly growing and expanding into various sectors, such as healthcare, transportation, smart homes, and more. Despite the benefits of using IoT devices, they present several challenges. Given the significant role these devices play in our lives, it is crucial to address issues related to their security and privacy. These devices are limited in resources, which complicates their security and the protection of the data that they manage. The paper aims to examine intrusion detection systems using the Gotham2025 dataset, generated through the Gotham testbed, which consists of 78 emulated IoT devices utilising various protocols, including MQTT, CoAP, and RTSP, to assist in safeguarding IoT networks from attacks. We conduct a comparative analysis between five machine learning algorithms, including Random Forest, XGBoost, Logistic Regression, Naive Bayes, and Deep Neural Network. We demonstrate that the Random Forest Classifier was the top-performing model, achieving an F1-score of 0.99 in classifying attacks.

In the evolving threat landscape, adversaries exploit software vulnerabilities to launch sophisticated attacks, challenging traditional defenses. Although databases like CVE and NVD provide detailed technical information, they often lack links to attacker behaviors such as tactics and techniques, limiting effective threat interpretation and response. This work bridges this gap by connecting vulnerabilities with behavioral patterns from the MITRE ATT&CK framework. We construct a CVE-TTP Knowledge Graph that links CVEs to tactics and techniques using classification and relation extraction. Transformer-based models are developed for behavior identification, with CySecBERT achieving macro F1-scores of 87.71% (techniques) and 96.16% (tactics). Also, we created an annotated dataset with 24,820 entities and 43,608 relations for entity and relation extraction. The pipeline-based approach achieves macro F1-scores of 0.86 (entity extraction) and 0.99 (relation extraction), while a span-based joint model achieves 0.78. These outputs are integrated into a Neo4j-based Cyber Threat Knowledge Graph, enabling structured visualization of vulnerabilities.

Certification of digital documents, such as academic credentials, seems a particularly suitable application for the use of blockchain and distributed ledger technologies. Indeed, these technologies enable decentralized certification systems that rely on the immutability and persistence of their distributed ledgers. However, in the absence of a central trusted authority, it is not easy to guarantee the authenticity of the connection between the real identity of an academic institution and the digital identity of the certificate issuer. In this paper, we demonstrate that one of such systems, known as Block.co, has a vulnerability that allows the production of forged certificates that are recognized as valid by the system. Since this is an inherent limitation of the approach used for blockchain-based certification, our attack is likely to be extendable to other systems adopting the same approach.

Generative image steganography synthesizes stego images directly from secret information to achieve inherent security advantages. Latent Diffusion Models (LDMs) have recently emerged as a fundamental image steganography framework that modulates secret latent representations with text prompts. Limited by the inflexibility of text prompts, these methods still struggle to generate high-quality stego images and accurately recover secret images. In this work, we propose a prompt-free diffusion image steganography framework that integrates style semantic priors to control more robust and reliable stego image generation. Specifically, a Cascaded Affine Coupling Module (CACM) establishes a bijective, deterministic mapping between a secret image and its latent representation. Then, style semantics are integrated into the diffusion process to control latent representation and ensure visual imperceptibility in the generated stego images. To mitigate trajectory deviations stemming from the unconditioned reverse process, a predictor-corrector mechanism is introduced to iteratively refine the generation trajectory via feedback from the current and predicted next states. Extensive experimental results show that the proposed method achieves competitive performance compared to state-of-the-art methods in terms of security, secret image reconstruction accuracy and controllability.

Large Language Models (LLMs) have rapidly proliferated, driving widespread adoption of AI applications. Most deployments rely on centralized infrastructures such as Microsoft Azure, Google Cloud, or AWS, requiring users to share sensitive data and training or fine-tuning code. This dependence raises significant security and privacy concerns, as cloud providers must be trusted to ensure confidentiality and integrity. Trusted Execution Environments (TEEs) e.g., Intel SGX/TDX, AMD SEV-SNP, and ARM CCA have been introduced to mitigate these risks. More recently, NVIDIA has developed GPU TEEs (e.g., H100/H200), yet comprehensive evaluations of end-to-end workflows that integrate CPU and GPU TEEs remain limited. Critical aspects, including performance overhead, remote attestation, and security guarantees for AI/LLM applications, have not been sufficiently studied. This paper addresses this gap by presenting an end-to-end workflow that combines CPU and GPU TEEs. We propose mechanisms to ensure confidentiality and integrity at both the VM level (via Intel TDX and AMD SEV-SNP) and the application level, highlighting vulnerabilities such as Kubernetes administrators' ability to access confidential VM contents. Finally, we evaluate the performance overhead of our system using industry benchmarks, focusing on configurations that integrate Intel TDX with NVIDIA H200 GPUs.

In cryptographic practice, where protocols impose strict time bounds, implementations demand predictable resource usage, and real-world systems require immediate verification for security and usability, a short key or certificate is useful only if it can be expanded or verified within a bounded time; otherwise a compact representation that requires superpolynomial work to expand offers no operational guarantee within a bounded-time protocol. This paper formalises that gap by introducing witness complexity \(\gam(x)\), the minimum running time over near-shortest descriptions of a string on a universal Turing machine. \(\gam\) differs from Shannon entropy and Kolmogorov complexity \(\KC\): low \(\KC\) can coexist with high \(\gam\). We prove invariance up to polynomial factors; a conditional separation (assuming \(\PneqNP\)). An unconditional lower bound from incomputability of \(\KC\); a biconditional characterisation of \(\PeqNP\) via the class-relative variant \(\gP\); and polynomial-time tractability for structured \(\classNP\) families. Part II develops companion measures and shows an unconditional gap between grammar size and derivation cost, positioning \(\gam\) as a metric for the usability of keys and certificates.

The quantitative record of science and engineering is increasingly carried by figures rather than text or tables, and a reader who needs the underlying numbers must usually re-digitize them by hand: slowly, imprecisely, and with no way to prove the result is faithful. Yet when a figure is stored as vector graphics, its data are not approximated by the picture but encoded in it: the renderer writes each marker and vertex at a printed precision that, for the dominant scientific toolchain, exceeds the data's own. We turn this into three contributions, one per shortcoming of hand digitization. First, a precision theory bounding how accurately data can be recovered for a given renderer and export format: bit-exact float32 for matplotlib markers, and a calibration-limited three to four significant figures end to end. Second, an automatic extractor that decodes a figure in one pass with no human in the loop, in place of the slow point-by-point tracing a digitizer demands. Third, a verification theory: recovery is injective except on a characterized, vanishingly small interval near zero; accidental agreement between unrelated data is astronomically unlikely; and a re-rendering certificate binds the recovered values to the markers, lines, and ticks the figure draws, not its text, making a result non-repudiable. With no ground truth used during recovery, decoded figures match external archives (Planck 2018 to 10^-9; the Keeling CO2 record to 5*10^-4, and one decoded figure independently reproduces a correction to the Chinchilla scaling-law confidence interval. We map the achievable precision across common renderers and their PDF, SVG, and EPS formats. What we deliver is certified data; the scientific significance of any particular dataset lies outside this paper's scope, and recovered values are candidates for human review, never accusations.

While post-training backdoor detection and trigger inversion schemes have been developed for AIs used e.g. for images, there is a paucity of such methods for LLMs. First, the LLM input space is discrete, with up to 150,000^k k-tuples to consider with k the token-length of a putative trigger. Second, one must blacklist tokens typical of the putative target response (class) of an attack, as such tokens may give false detection signals. However, a comprehensive blacklist is not available, in general, for a given domain. We develop a highly effective detection and inversion framework for LLMs treated as classifiers. Central to our approach is class subspace orthogonalization (CSO), a novel plug-and-play paradigm for backdoor detection that serves two fundamental roles when applied to LLMs: i) it enhances both sensitivity and specificity of a baseline detector; ii) it provides a form of implicit blacklisting, as it penalizes against inclusion, in a candidate trigger, of tokens that induce signal perturbations "in the direction of" the putative target class of an attack. One version of our detector performs continuous optimization in token embedding space, while a companion trigger-inversion and detection method performs greedy accretion in discrete token space. Our methods give both strong detection performance and accurate inversion of ground-truth triggers on several LLM classification domains, and for several different LLM architectures.

AI agents increasingly acquire and execute skills at runtime: bundles of prompt instructions, executable code, and tool declarations fetched from marketplaces and other agents. Governing them needs a stable notion of skill identity, yet cryptographic hashing is engineered to destroy the very similarity we need, as a one-character edit scrambles the digest. We present a compact, locality-sensitive fingerprint that embeds each component of a skill and projects it to bits with a multi-bank SimHash, giving a fixed 120-byte signature compared in constant time by Hamming distance. Our central claim is that keeping the fingerprint as a per-component triple (prompt, code, tools), rather than a single score, is what makes it useful: the triple recovers skill-family identity through paraphrase, renaming, refactoring, and controlled code translation when another component remains shared, while independent multilingual reimplementation is not recovered; it also localizes which component carries the reuse. We claim lineage, not behavioral equivalence: identity supplies the structural axis of a registry and leaves safety to behavioral verification. The fingerprint reaches an area under the ROC curve (AUC) of 0.974 (95% CI [0.956, 0.994]) over 4,950 pairwise comparisons while using 77x fewer bits than the embedding it approximates, with ranking preserved in expectation and finite-bit concentration; the per-component split turns one number into relationship classification, families, novelty, and a portable "SkillBOM" for a skill registry. On a 906-skill injection benchmark the fingerprint recognizes injected skills as tampered copies of a known base and localizes the change, but recognition is not trust: it remains, by design, an identity signal complementary to behavioral verification rather than a safety verdict.

The fast growth of open-source AI infrastructure, from model serving engines and agent platforms to the Model Context Protocol (MCP) ecosystem and the language models themselves, has outpaced the security tooling available to defend it. We present AI-Infra-Guard, an open-source framework that organizes AI red teaming around a single observation: the attack surface of an AI agent is stratified across layers (infrastructure, protocol/tool, agent behavior, and model), and no single detection paradigm fits all of them. The framework therefore matches a paradigm to each layer, from deterministic rule matching over 75+ AI components and 1{,}400+ vulnerability rules, through LLM-driven agentic auditing of MCP servers and agent-skill packages and multi-turn black-box agent red teaming, to a jailbreak harness with 26+ attack operators over sixteen datasets. To our knowledge it is the only open-source framework to span all of these, including supply-chain auditing of the agent skills that increasingly extend AI agents. We release AI-Infra-Guard as open source so that layer-paradigm matching can serve as a practical foundation for agent security and a shared base for the community to build on.

Large tabular models (LTMs), i.e., tabular foundation models leveraging in-context learning (ICL), achieve state-of-the-art performance on tabular tasks. While LLMs are known to unintentionally memorize training data, the memorization dynamics of LTMs remain largely unexplored. We investigate the potential for parametric memorization in tabular ICL. We introduce ICLMEM, a probing framework designed to separate context-based predictions from parametric memorization. Our zero-information multiple-choice context strips away valid contextual patterns to force the model to fall back on its parametric memory. Our controlled fine-tuning setup establishes membership ground truth and accounts for common pitfalls, e.g., distribution shift, feature contamination, base-rate fallacy, and the pre-trained base model acts as reference to calibrate for sample difficulty. Our controlled evaluation on a leading real-world-trained LTM detects moderate memorization signals in 8 out of 10 tasks ($\text{AUC}$ up to $0.67$ and TPR at $1\%$ FPR $>0.1$). Notably, memorization signals are strongest for low-cardinality and binary tasks. However, they largely vanish under realistic training conditions. Our findings show LTM memorization signals under specific circumstances (single-task fine-tuning with fixed samples across many epochs and small query size). To protect sensitive data, appropriate measures must be taken, which we discuss.

We audit a fixed prefix-window mean-NLL memorization probe (K=20) on a Qwen2.5-VL-7B canary testbed and report three post-hoc cases where it disagrees with full-span secret NLL or greedy exact-recall. C3 (false negative, window truncation): damage lands on hex tokens outside K=20; the probe stays flat while hit@1 drops. C4 (false positive, non-secret drift): the probe moves, but approximately 99% sits on non-secret preamble; the secret span and hit@1 are unchanged. C5 (ambiguous in-window drop): the probe falls on an undertrained baseline while full-span hex is positive and hit@1=0. Recommendation: report (i) full-span secret NLL, (ii) a span-localised decomposition, (iii) behavioural exact-recall at k>=4, and (iv) decoy probes before asserting secret-specificity. Evidence is on controlled canaries in one backbone; magnitudes are testbed-specific.

Large Language Models (LLMs) are rapidly transforming software development, yet their use in security-critical contexts raises a key question: do models know when their generated code is insecure? This property, known as calibration, measures whether a model's confidence aligns with the true correctness of its outputs. We present the first large-scale empirical study of security calibration in LLM-generated code. We evaluate GPT-4o-mini, Gemini-2.0-Flash, and Qwen3-Coder-Next across multiple temperature settings on two complementary benchmarks: self-contained security tasks and multi-language repository-level contexts. Our results suggest that overconfidence is prevalent across the evaluated LLMs. Functional calibration is consistently worse than security calibration, suggesting that models estimate security outcomes more reliably than functional correctness, potentially because functional correctness depends on complex execution behavior. We also examine whether calibration-guided automated repair can help remediate vulnerabilities in LLM-generated code, finding only limited improvements while frequently introducing functional regressions. Moreover, we study different mitigation strategies for reducing False Trust, where models assign high confidence to vulnerable code. The results show that although architectural gating improves calibration on controlled benchmarks, calibration deteriorates in realistic repository-level settings, increasing the risk of high-confidence vulnerable outputs.

Federated Learning (FL) is highly susceptible to stealthy backdoor attacks, which aim to force a model into predicting an attacker-chosen target class for inputs containing a specific trigger. However, existing statistical defenses primarily focus on the early stages of model convergence. In this paper, we identify a fundamental vulnerability termed “Late-stage Failure.” We demonstrate that as the global model converges, decaying gradient norms render malicious and benign updates morphologically indistinguishable. This vanishing statistical variance effectively blinds traditional defenses, enabling adaptive adversaries to remain dormant and subsequently hijack the training process. To overcome these constraints, we propose Secure-CHG, a hybrid framework that pivots the defense paradigm from superficial morphological detection toward intrinsic semantic contribution verification. Secure-CHG employs an adaptive defense pipeline: a cascaded statistical filter stabilizes optimization during the early oscillatory phase, while a novel CHG-Shapley mechanism takes over during late-stage convergence. By leveraging sample hardness (i.e., local training loss) to project updates into a composite Hardness-Gradient space, it effectively amplifies adversarial semantic traces, enabling the isolation of stealthy attackers even as gradient norms vanish. Furthermore, we derive a closed-form solution for CHG-Shapley, facilitating low-complexity, retraining-free node valuation and trust-modulated aggregation. Extensive evaluations on CIFAR-10, MedMNIST, and NEU-SDDB demonstrate that Secure-CHG effectively mitigates Late-stage Failure. Specifically, it significantly suppresses advanced backdoor attacks, reducing their attack success rate by 2.3$\times$ and 2.0$\times$ relative to the mainstream Krum and Trimmed Mean baselines, respectively.

Hard-constrained sequential decision systems have no certified way to spend the test-time compute of modern AI: executing the multi-step drafts of a learned policy or a frozen LLM forfeits the feasibility guarantee a trusted solver provides, while invoking the solver at every step forfeits the speed the AI offers. Certificate-Gated Prefix Acceptance (CGPA) closes this gap with a certified speculative-execution contract for untrusted AI agents: a trusted verifier rejects constraint-violating transitions exactly, a conformally calibrated value boundary gates the longest low-cost prefix within a per-segment regret budget, and the rest defers to the solver, so safety, regret, and speed decouple by construction. The contract drives every untrusted proposal source - adversarial drafters and six heterogeneous frozen LLMs (including a 12B model that violates constraints in 98% of direct rollouts) - to zero applied violations; a certificate-aware learned boundary, conformally calibrated, drives mean regret three orders of magnitude below unguarded acceptance, to within sampling noise of the stepwise oracle (95% CI spanning zero), and under calendar shift a learned proposal source overtakes it on 15 of 18 held-out days. On a deployment-scale unit-commitment instance it turns a frozen 8B LLM into a 2.96x per-episode wall-clock speedup at 2.1% regret, outpacing the domain heuristic (1.79x) and a safe receding-horizon baseline (1.07x): the more capable the untrusted source, the faster the certified system, at guarantees that never change.

Backdoor attacks pose a serious threat to large language models (LLMs) by causing otherwise benign systems to produce attacker-specified malicious behavior when a hidden trigger is present. In this work, we study post hoc detoxification of backdoored LLMs in a practical setting where the defender has access to the poisoned model but does not wish to retrain the full network from scratch. We propose a mechanistically guided weight-space repair framework that first localizes modules involved in propagating trigger-induced behavior using activation patching and Fisher/K-FAC curvature analysis, and then applies targeted low-rank repair to only the most influential modules. We evaluate the method on poisoned variants of Llama-3.2-1B-Instruct with triggers inserted at the beginning, middle, and end of otherwise benign prompts. Results show that the proposed approach substantially suppresses trigger-conditioned malicious responses while preserving benign model behavior. These findings suggest that backdoor removal in LLMs can be formulated as a localized structural repair problem rather than only a broad behavioral alignment problem.

Generative AI has emerged as a significant cybersecurity threat, with several recent attack campaigns leveraging LLMs to generate code for malicious purposes via scripting languages such as PowerShell. Consequently, for cybersecurity analysts, it is imperative to investigate the offensive capabilities of AI code generators. In this paper, we propose an experimental framework to assess LLM-generated PowerShell malware, which comprises a novel sandbox approach for dynamic analysis of AI-generated malware. Furthermore, we present a novel, manually curated dataset of real-world PowerShell malware, annotated in natural language to assist the training and evaluation of LLMs. Finally, this study evaluates permissive, open-weight LLMs adapted to PowerShell malware generation. Our results reveal a high degree of similarity between real malware and LLM-generated ones in terms of triggered OS malicious events, with a median Jaccard index of 84.5% and 48.4% of instances achieving complete overlap.

Generative models have recently seen rapid adoption in End-to-End (E2E) autonomous driving (AD), with diffusion-based denoising and vocabulary-based retrieval becoming the dominant trajectory-decoding paradigms. Despite their architectural diversity, current generative AD planners share a common inference pattern: a fixed set of candidate trajectories (anchors, vocabulary entries, or proposal queries) is scored by one or more learned heads conditioned on the Bird's-Eye-View (BEV) features, and the highest-scored candidate is returned as the final trajectory. Under this design, the scoring head is the only barrier between perception and the motion command, and its decision margins between competing candidates are often small. We introduce Derail, an adversarial framework that exploits this scoring-head attack surface. Evaluated on various generative planners, Derail flips the trajectory selection from a safe to an unsafe candidate, with score drops of $39$–$80\%$ and collision rates of up to $50\%$, consistently outperforming generic loss-maximization and feature-divergence attacks. Our analysis suggests that safety-violating objectives govern attack effectiveness against generative AD planners, and that the scoring-head inference pattern itself is a recurring attack surface worth explicit defensive consideration.

Language models are often adapted in stages: a public skill phase, a private memory phase, and a later safety phase that learns to refuse outputs tied to the remembered entities. Revoking the memory after the safety phase is not the same problem as subtracting the memory update: the later safety optimizer has transported the memory direction. We introduce process sidecars, a two-coefficient edit family $\hat{\theta}(\lambda,\gamma)=\theta_{\mathrm{AMS}}-\lambda\Delta_{\mathrm{M}}-\gamma\hat{R}_{\mathrm{S}\leftarrow\mathrm{M}}$, with $\hat{R}_{\mathrm{S}\leftarrow\mathrm{M}}=\hat{J}_{\mathrm{S},\varepsilon}(\Delta_{\mathrm{M}})-\Delta_{\mathrm{M}}$, where $\hat{J}_{\mathrm{S},\varepsilon}$ is a centered secant through the realized future AdamW safety-training process. The implementation uses $\varepsilon=1$ at the natural memory-edit scale; it reuses $\theta_{\mathrm{AMS}}$ as the positive endpoint and computes one additional safety trace at $\theta_{\mathrm{A}}-\Delta_{\mathrm{M}}$. We prove two things. First, the exact sidecar, using the true transported direction $R_{\mathrm{S}\leftarrow\mathrm{M}}$ rather than the secant estimate, at $(\lambda,\gamma)=(1,1)$ recovers the counterfactual safety-only oracle $\theta_{\mathrm{AS}}$ up to second order; the proof treats AdamW as an augmented-state map over parameters, first moments, and second moments. Second, this process information is necessary: whenever future safety training bends the memory direction, every scalar task-arithmetic edit leaves first-order counterfactual error, while the process-sidecar edit is second-order accurate. Across three models, the validation-selected 2D edit improves held-out refusal closure over naive task arithmetic in all trials, and over the $\gamma=\lambda$ process-JVP subfamily, the diagonal slice of the cached 2D grid, in all paired trials.

We identify a security-fidelity tradeoff in defending LLMs against indirect prompt injection: defenses resist injected instructions largely by suppressing untrusted text, which corrupts tasks that must preserve it, such as translation and document editing. Attack-success metrics cannot see this, because a model that ignores an injection and one that faithfully processes it as data score identically. We introduce SecFid, a benchmark built so that executing an injection, processing it as data, and ignoring it produce distinguishable outputs. This makes fidelity measurable and exposes a frontier: across 1,168 examples and 48 configurations, no model or defense achieves both objectives. The highest-fidelity model reaches 96.5% fidelity at 47.8% security, while the most secure defenses invert this, at 99.3% security but only 71.0%-73.9% fidelity. Even defenses with identical security differ in how they earn it: some repair hijacks into faithful processing, others simply suppress benign content. A decision-theoretic analysis shows why no fixed choice can be right everywhere: the correct behavior is not a property of the defense but of the deployment, set by its relative cost of a hijack versus a dropped span. Security alone therefore measures only half of robustness, and reporting it without fidelity hides the price at which it was bought.

Claw-like AI agents (e.g., OpenClaw) are always-on processes with persistent access to credentials, files, tools, and external services. They take on system-level responsibilities – installing packages, maintaining state, scheduling subtasks, and mediating I/O – making security failures far more severe than in other agents. Yet existing benchmarks focus on model responses and tool calls, leaving cross-component failure modes largely unmeasured. We adopt a computer-system analogy: treating a Claw-like agent as an agentic computer system whose gateway runtime plays an OS-like mediation role, whose Skills resemble user-installed applications, and whose Plugins resemble loadable extensions with runtime privileges. Each component has a classical counterpart whose protection mechanisms – refined over decades of cybersecurity research – are absent on the agent side. From this perspective, we develop SafeClawArena, a benchmark of 406 adversarial tasks across four attack surfaces (Skill Supply-Chain Integrity, Persistent State Exploitation, Cross-Boundary Data Flow, and Indirect Prompt Injection), executed in containerized replicas of real agent platforms with canary-marked credentials and evaluated via automated taint tracking across nine output channels. We evaluate three platforms (OpenClaw, NemoClaw, SeClaw) and five frontier LLMs. The highest attack success rate reaches 70%; malicious Plugins succeed in 100% of cases regardless of the LLM. SeClaw cuts GPT-5.4's attack success rate from 70% to 22%, partly through utility-security tradeoffs rather than active defenses, while Claude-Opus-4.6 already sits near a 22% floor on every platform. These results expose the inadequacy of current defenses and suggest directions for future hardening. Code and data: https://github.com/sunblaze-ucb/SafeClawArena.

As the Internet of Things (IoT) continues its rapid expansion, the attack surface grows accordingly, with emerging threats targeting smart objects and their interactions. In this evolving landscape, securing service provisioning is crucial to ensure the proper functioning, security, and reliability of the IoT ecosystem. Service provisioning encompasses key tasks such as device registration, configuration, authentication, authorization, and software deployment, all of which are essential for seamless and secure IoT operations. In this paper, we present a comprehensive framework designed to select the most suitable smart objects to deliver a target service within a given IoT environment while also monitoring the behavior of the entities involved during the service provisioning phase. To achieve this, we employ a Deep Reinforcement Learning (DRL) approach in which an intelligent agent learns, through interaction with a complex, dynamic environment, how to adapt to changes while adhering to predefined security constraints. For behavioral monitoring, we leverage Federated Learning (FL) to develop a global Behavioral Fingerprinting (BF) model that is fully distributed and can analyze how IoT devices interact within the network. In addition, the BF is used to compute a reliability score for each service provider, reflecting its degree of compliance with the defined security constraints. This score is then incorporated into the service provisioning process, allowing smart objects to select providers not only according to functional suitability but also to their reliability level. Finally, we conduct an extensive experimental evaluation to assess the robustness and scalability of our approach. The results demonstrate that our solution can be effectively deployed even on resource-constrained IoT devices, making it a viable and scalable security-enhancing mechanism for modern IoT ecosystems.

In this review, we survey the cryptographic task of authentication from the perspective of quantum communication. We review three main flavours of authentication that are often conflated in the literature: authentication of classical messages, authentication of quantum messages, and entity authentication, also covering recent hardware-assisted approaches. We compare representative protocols for each functionality in terms of their security assumptions, set-up requirements, composability, and scalability in large or dynamic networks, and use these criteria to identify and recommend suitable candidates. Finally, applications are surveyed: we provide a detailed case study of authentication and quantum key distribution (QKD), then extend the discussion to protocols beyond QKD, where the role of authentication is more complex. Our take-home message is that an authentication requirement is not an intrinsic limitation of quantum networks: as with all secure communication, each protocol relies on a particular authentication resource, and the security claim of that protocol is meaningful only once the authentication resource and its deployment assumptions are made explicit. At the same time, the existing classical and quantum literature already offers a range of quantum-secure authentication schemes, which can support different applications when carefully matched to the required functionality, assumptions, and security guarantees.

Multi-agent systems (MAS) are increasingly used to automate complex, distributed workflows. However, their inter-agent communication channels introduce new attack surfaces that remain poorly understood and are difficult to defend against. In this paper, we address how defenders should prioritize limited security effort to protect vulnerable communication channels before attacks are observed. This is motivated by our observation that the channel-level attack impact is highly non-uniform: a single compromised edge can account for up to 75% of total attack success. We introduce Mesa, a label-free framework for proactively ranking which MAS edges are most security-critical – that is, most likely to affect the system's decision if compromised. Mesa combines six graph-theoretic metrics and two dynamic probes (ablation and masking) without requiring attack traces. We evaluate Mesa against a dynamic misinformation attack pipeline across three diverse MAS scenarios, eight network topologies, and five open-source LLMs from Qwen, Llama, and Gemma families. Mesa rankings correlate strongly with empirical per-edge attack success rate, achieving mean Spearman $\rho=+0.60$ (peaking at $+0.73$). In resource-constrained defense deployment, monitoring the top 10% of Mesa-ranked edges intercepts about 3x the successful attacks as random allocation. We further test Mesa under varying attacker and defender models and LangGraph workflows and characterize its limits under adaptive attacks and high-redundancy graphs. Overall, our results show that edge-level risk in MAS is often concentrated and predictable, allowing proactive hardening of multi-agent infrastructures.