Papers

Large Language Models (LLMs) show remarkable capabilities in understanding natural language and generating complex code. However, as practitioners adopt CodeLLMs for increasingly critical development tasks, research reveals that these models frequently generate functionally correct yet insecure code, posing significant security risks. While multiple approaches have been proposed to improve security in AI-based code generation, combined benchmarks show these methods remain insufficient for practical use, achieving only limited improvements in both functional correctness and security. This stems from a fundamental gap in understanding the internal mechanisms of code generation and the root causes of security vulnerabilities, forcing researchers to rely on heuristics and empirical observations. In this work, we investigate the internal representation of security concepts in CodeLLMs, revealing that models are often aware of vulnerabilities as they generate insecure code. Through systematic evaluation, we demonstrate that CodeLLMs can distinguish between security subconcepts, enabling a more fine-grained analysis than prior black-box approaches. Leveraging these insights, we propose Secure Concept Steering for CodeLLMs (SCS-Code). During token generation, SCS-Code steers LLMs' internal representations toward secure and functional code output, enabling a lightweight and modular mechanism that can be integrated into existing code models. Our approach achieves superior performance compared to state-of-the-art methods across multiple secure coding benchmarks.

Network intrusion detection systems play a crucial role in the security strategy employed by organisations to detect and prevent cyberattacks. Such systems usually combine pattern detection signatures with anomaly detection techniques powered by machine learning methods. However, the commonly proposed machine learning methods present drawbacks such as over-reliance on labeled data and limited generalization capabilities. To address these issues, embedding-based methods have been introduced to learn representations from network data, such as DNS traffic, mainly due to its large availability, that generalise effectively to many downstream tasks. However, current approaches do not properly consider contextual information among DNS queries. In this paper, we tackle this issue by proposing DNS-GT, a novel Transformer-based model that learns embeddings for domain names from sequences of DNS queries. The model is first pre-trained in a self-supervised fashion in order to learn the general behavior of DNS activity. Then, it can be finetuned on specific downstream tasks, exploiting interactions with other relevant queries in a given sequence. Our experiments with real-world DNS data showcase the ability of our method to learn effective domain name representations. A quantitative evaluation on domain name classification and botnet detection tasks shows that our approach achieves better results compared to relevant baselines, creating opportunities for further exploration of large-scale language models for intrusion detection systems. Our code is available at: https://github.com/m-altieri/DNS-GT.

For each prime $p$, let $c(p)=\frac{\varphi(p-1)}{p-1}\prod_{j\ge1}(1-p^{-j})$, the limiting probability, as $n\to\infty$, that a uniformly chosen matrix in $M_n(\mathbb{F}_p)$ has primitive-root determinant. As $p$ ranges over the primes, $c(p)$ has the classical shifted-totient limiting law $\mu_G$ on $[0,1/2]$. Let $\mu_f$ be its image under $x\mapsto-\log x$. We prove that $\mu_G$ has Hausdorff dimension zero and gives full mass to absolutely normal Liouville numbers. Every Borel set of positive $\mu_G$-mass has packing dimension at least $1/2$, and local dimension fails to exist $\mu_G$-almost everywhere. Both Fourier transforms satisfy the effective bound $O((\log|\tau|)^{-1/3}(\log\log|\tau|)^{1/3})$. In every fixed reduced progression, the empirical laws of $c(p)$ and $\varphi(p-1)/(p-1)$ approach their respective limiting laws with sharp Kolmogorov order $1/\log x$. Over unrestricted primes, these laws and their negative-logarithmic images converge in $W_1$ at rate $O_B((\log x)^{-B})$ for every fixed $B>0$. We obtain concentration expansions to every fixed inverse-logarithmic order and prove that every sufficiently short maximizing interval for $\mu_f$ contains $\log3$. For $U_T$ uniform on $[0,T]$, $(\log T)\widehat{\mu_f}(U_T)$ converges in distribution to a circularly symmetric random variable whose modulus is an explicit random Euler product, with convergence of absolute moments of every real order $s>-1$. For every $A>0$, the strongly additive function $f_A$ with prime values $(\log\ell)^{-A}$ has absolutely continuous limiting laws for $f_A(n)$ and $f_A(p-1)$. We give extremal estimates and certified NTT prime searches. A lattice supplement determines second distinct nonzero squared norms of type-$A$ tensor products and gives uniform cyclotomic smoothing asymptotics at $\epsilon=2^{-c\varphi(m)}$ for every fixed $c>0$. These statements are unconditional.

Large language models remain vulnerable to jailbreak attacks, yet we still lack a systematic understanding of how jailbreak success scales with attacker effort across methods, model families, and harm types. We initiate a scaling-law framework for jailbreaks by treating each attack as a compute-bounded optimization procedure and measuring progress on a shared FLOPs axis. Our systematic evaluation spans four representative jailbreak paradigms, covering optimization-based attacks, self-refinement prompting, sampling-based selection, and genetic optimization, across multiple model families and scales on a diverse set of harmful goals. We investigate scaling laws that relate attacker budget to attack success score by fitting a simple saturating exponential function to FLOPs–success trajectories, and we derive comparable efficiency summaries from the fitted curves. Empirically, prompting-based paradigms tend to be the most compute-efficient compared to optimization-based methods. To explain this gap, we cast prompt-based updates into an optimization view and show via a same-state comparison that prompt-based attacks more effectively optimize in prompt space. We also show that attacks occupy distinct success–stealthiness operating points with prompting-based methods occupying the high-success, high-stealth region. Finally, we find that vulnerability is strongly goal-dependent: harms involving misinformation are typically easier to elicit than other non-misinformation harms.

In a content delivery network (CDN), resources are strained during peak-time and underutilised in off-peak times when supplying digital content to users. Caching can help balance this. At the off-peak time some content is delivered to users' local caches. During peak time, the use of cached data to serve users' requests relieves strain on the network by reducing repeated transfer of popular content. In coded caching, the cache content placement is designed in conjunction with the delivery techniques to optimise network throughput. Since dissemination of information, as well as the delivery of entertainment, is reliant on CDNs, the security and privacy of cache placement, user demand, and content delivery, are paramount. In much of the literature in secure coded caching, security is built on top of solutions that have efficiency in mind, and most current proposals focus on the security of individual parts of the process. A lack of a unifying network model also makes it difficult to compare or combine solutions. In this survey we analyse the security and privacy requirements of secure coded caching, and evaluate existing schemes in terms of the security provided and the cost of this security provision. We also review the techniques used to achieve secure coded caching and analyse their limitations. In addition, we contextualise secure coded caching in the landscape of other secure content delivery primitives. As a result, we identify and prioritise open security and privacy challenges for the future.

Communication topology is a critical factor in the utility and safety of LLM-based multi-agent systems (LLM-MAS), making it a high-value intellectual property (IP) whose confidentiality remains insufficiently studied. Existing topology inference attempts rely on impractical assumptions, including control over the administrative agent and direct identity queries via jailbreaks, which are easily defeated by basic keyword-based defenses. As a result, prior analyses fail to capture the real-world threat of such attacks. To bridge this realism gap, we propose WebWeaver, an attack framework that infers the complete LLM-MAS topology by compromising only a single arbitrary agent instead of the administrative agent. Unlike prior approaches, WebWeaver relies solely on agent contexts rather than agent IDs, enabling significantly stealthier inference. WebWeaver further introduces a new covert jailbreak-based mechanism and a novel fully jailbreak-free diffusion design to handle cases where jailbreaks fail. Additionally, we address a key challenge in diffusion-based inference by proposing a masking strategy that preserves known topology during diffusion, with theoretical guarantees of correctness. Extensive experiments show that WebWeaver substantially outperforms state-of-the-art (SOTA) baselines, achieving about 60% higher inference accuracy under active defenses with negligible overhead.

AI agents that combine large language models with non-AI system components are rapidly emerging in real-world applications, offering unprecedented automation and flexibility. However, this unprecedented flexibility introduces complex security challenges fundamentally different from those in traditional software systems. This paper presents the first systematic and comprehensive survey of AI agent security, including an analysis of the design space, attack landscape, and defense mechanisms for secure AI agent systems. We further conduct case studies to point out existing gaps in securing agentic AI systems and identify open challenges in this emerging domain. Our work also introduces the first systematic framework for understanding the security risks and defense strategies of AI agents, serving as a foundation for building both secure agentic systems and advancing research in this critical area.

We resolve an open question of Jain, Raskhodnikova, Sivakumar, and Smith (ICML 2023) by exhibiting a problem separating differential privacy under continual observation in the oblivious and adaptive settings. The continual observation (a.k.a. continual release) model formalizes privacy for streaming algorithms, where data is received over time and output is released at each time step. In the oblivious setting, privacy need only hold for data streams fixed in advance; in the adaptive setting, privacy is required even for streams that can be chosen adaptively based on the streaming algorithm's output. We describe the first explicit separation between the oblivious and adaptive settings. The problem showing this separation is based on the correlated vector queries problem of Bun, Steinke, and Ullman (SODA 2017). Specifically, we present an $(\varepsilon,0)$-DP algorithm for the oblivious setting that remains accurate for exponentially many time steps in the dimension of the input. On the other hand, we show that every $(\varepsilon,\delta)$-DP adaptive algorithm fails to be accurate after releasing output for only a constant number of time steps.

In this paper, we present a laboratory study focused on the impact of post-quantum cryptography (PQC) algorithms on multiple layers of stateful HTTP over TLS transactions: the TCP handshake, the intermediate TCP-TLS layer, the TLS handshake, the intermediate TLS layer, and the HTTP application layer. To this end, we propose a laboratory architecture that emulates a real-world setup in which a load test of up to 100 transactions per second is sent to a load balancer, which in turn forwards them to a backend server that returns the responses. Each set of tests is executed using the TLS 1.3 key exchange groups as follows: traditional (or non-PQC), hybrid PQC and pure PQC. Each set of tests also varied the backend response size. Across more than thirty experiments, we performed data reduction and statistical analysis for each layer, to determine the specific impact of each algorithm (PQC and traditional) at every stage of the HTTP-over-TLS transaction.

With their increasing capabilities, Large Language Models (LLMs) are now used across many industries. They have become useful tools for software engineers and support a wide range of development tasks. As LLMs are increasingly used in software development workflows, a critical question arises: are LLMs good at software security? At the same time, organizations worldwide invest heavily in cybersecurity to reduce exposure to disruptive attacks. The integration of LLMs into software engineering workflows may introduce new vulnerabilities and weaken existing security efforts. We introduce TOSSS (Two-Option Secure Snippet Selection), a benchmark that measures the ability of LLMs to choose between secure and vulnerable code snippets. Existing security benchmarks for LLMs cover only a limited range of vulnerabilities. In contrast, TOSSS relies on the CVE database and provides an extensible framework that can integrate newly disclosed vulnerabilities over time. Our benchmark gives each model a security score between 0 and 1 based on its behavior; a score of 1 indicates that the model always selects the secure snippet, while a score of 0 indicates that it always selects the vulnerable one. We evaluate 14 widely used open-source and closed-source models on C/C++ and Java code and observe scores ranging from 0.48 to 0.89. LLM providers already publish many benchmark scores for their models, and TOSSS could become a complementary security-focused score to include in these reports.

Vulnerabilities emanating from DRAM errors pose a vexing problem that remains, as of yet, unsolved and elusive but cannot be ignored. Prior defenses focused on specific details of early RowHammer attacks and fail to generalize with the generalizations of recent RowHammer attacks. Even worse, it is presently not clear that techniques from prior defenses will be able to cope with these generalizations or if an entirely new approach is required. Although still work-in-progress, we have identified a new approach that combines memory allocation with principles underlying software diversity and shows promising early results. At first glance, software diversity seems to be an unlikely contender, since it faces seemingly insurmountable obstacles, primarily the lack of sufficient entropy in memory subsystems. Our system - called MAD, short for memory allocation diversity - leverages two novel, complementary spatial diversification techniques to overcome this entropy obstacle. Entropy aside, MAD offers ease-of-implementation, negligible performance impact, and is both hardware and software agnostic. From a security perspective, MAD's goal is to deter RowHammer attacks by delaying them to the maximum extent possible. Such a delay opens the door for a variety of additional responses, e.g., proactive rebooting, or complementary in-depth analysis of ongoing attacks that would be too slow for an always-on defense.

This paper investigates how Backdoor Attacks are represented within Vision Transformers (ViTs). By assuming knowledge of the trigger, we identify a specific “trigger direction” in the model's activations that corresponds to the internal representation of the trigger. We confirm the causal role of this linear direction by showing that interventions in both activation and parameter space consistently modulate the model's backdoor behavior across multiple datasets and attack types. Using this direction as a diagnostic tool, we trace how backdoor features are processed across layers. Our analysis reveals distinct qualitative differences: static-patch triggers follow a different internal logic than stealthy, distributed triggers. We further examine the link between backdoors and adversarial attacks, specifically testing whether PGD-based perturbations (de-)activate the identified trigger mechanism. Finally, we propose a data-free, weight-based detection scheme for stealthy-trigger attacks. Our findings show that mechanistic interpretability offers a robust framework for diagnosing and addressing security vulnerabilities in computer vision.

EVMbench, released by OpenAI, Paradigm, and OtterSec, is the first large-scale benchmark for AI agents on smart contract security. Its results – agents detect up to 45.6% of vulnerabilities and exploit 72.2% of a curated subset – have fueled expectations that fully automated AI auditing is within reach. We identify two limitations: its narrow evaluation scope (14 agent configurations, most models tested on only their vendor scaffold) and its reliance on audit-contest data published before every model's release that models may have seen during training. To address these, we expand to 26 configurations across four model families and three scaffolds, and introduce a contamination-free dataset of 22 real-world security incidents postdating every model's release date. Our evaluation yields three findings: (1) agents' detection results are not stable, with rankings shifting across configurations, tasks, and datasets; (2) on real-world incidents, no agent succeeds at end-to-end exploitation across all 110 agent-incident pairs despite detecting up to 65% of vulnerabilities, contradicting EVMbench's conclusion that discovery is the primary bottleneck; and (3) scaffolding materially affects results, with an open-source scaffold outperforming vendor alternatives by up to 5 percentage points, yet EVMbench does not control for this. These findings challenge the narrative that fully automated AI auditing is imminent. Agents reliably catch well-known patterns and respond strongly to human-provided context, but cannot replace human judgment. For developers, agent scans serve as a pre-deployment check. For audit firms, agents are most effective within a human-in-the-loop workflow where AI handles breadth and human auditors contribute protocol-specific knowledge and adversarial reasoning. Code and data: https://github.com/blocksecteam/ReEVMBench/.

The expansion of Internet of Things (IoT) devices has increased the attack surface of networks, necessitating a robust and adaptive intrusion detection systems. Machine learning based systems have been considered promising in enhancing the detection performance. Federated learning settings enabled us to train models from network intrusion data collected from clients in a privacy preserving manner. However, the effectiveness of these systems can degrade over time due to concept drift, where patterns in data evolve as attackers develop new techniques. Realistic detection models should be non-stationary, so they can be continuously updated with new intrusion data while maintaining their detection capability for older data. As IoT environments are resource constrained, updates should consume minimal computational resources. This study provides a comprehensive performance analysis of incremental federated learning in enhancing the long term performance of non stationary IDS models in IoT networks. Specifically, we propose LSTM models within a federated learning setting to evaluate incremental learning approaches that utilize data and model-based measures against catastrophic learning under drift conditions. Using the CICIoMT2024 dataset, which includes various attack variants across five major categories, we conduct both binary and multiclass classification to provide a granular analysis of the intrusion detection task. Our results show that cumulative incremental learning and representative learning provide the most stable performance under drift, while retention-based methods offer a strong accuracy and latency trade off. The study offers new insights into the interplay between training strategy performance and latency in dynamic IoT environments, aiming to inform the development of more resilient IDS solutions considering the resource constraints in IoT devices.

More and more companies' Intellectual Property (IP) is being integrated into Neural Network (NN) models. This IP has considerable value for companies and, therefore, requires adequate protection. For example, an attacker might replicate a production machines' hardware and subsequently simply copy associated software and NN models onto the cloned hardware. To make copying NN models onto cloned hardware infeasible, we present an approach to bind NN models - and thus also the IP contained within them - to their underlying hardware. For this purpose, we link an NN model's weights, which are crucial for its operation, to unique and unclonable hardware properties by leveraging Physically Unclonable Functions (PUFs). By doing so, sufficient accuracy can only be achieved using the target hardware to restore the original weights, rendering proper execution of the NN model on cloned hardware impossible. We demonstrate that our approach accomplishes the desired degradation of accuracy on various NN models and outline possible future improvements.

LLM agents are highly vulnerable to Indirect Prompt Injection (IPI), where adversaries embed malicious directives in untrusted tool outputs to hijack execution. Most existing defenses treat IPI as an input-level semantic discrimination problem, which often fails to generalize to unseen payloads. We propose a new paradigm, action-level causal attribution, which secures agents by asking why a particular tool call is produced. The central goal is to distinguish tool calls supported by the user's intent from those causally driven by untrusted observations. We instantiate this paradigm with AttriGuard, a runtime defense based on parallel counterfactual tests. For each proposed tool call, AttriGuard verifies its necessity by re-executing the agent under a control-attenuated view of external observations. Technically, AttriGuard combines teacher-forced shadow replay to prevent attribution confounding, hierarchical control attenuation to suppress diverse control channels while preserving task-relevant information, and a fuzzy survival criterion that is robust to LLM stochasticity. Across four LLMs and two agent benchmarks, AttriGuard achieves 0% ASR under static attacks with negligible utility loss and moderate overhead. Importantly, it remains resilient under adaptive optimization-based attacks in settings where leading defenses degrade significantly.

Large Language Models (LLMs) rely on optimizations like Automatic Prefix Caching (APC) to accelerate inference. APC works by reusing previously computed states for the beginning part of a request (prefix), when another request starts with the same text. While APC improves throughput, it introduces timing side channels: cache hits are faster than misses, creating observable latency differences. In multi-tenant systems, attackers can exploit these differences to infer sensitive information, e.g., by incrementally reconstructing another user's request by observing hit/miss patterns. Current defenses take a sledgehammer approach: they disable APC and cache sharing, isolating users, and sacrificing efficiency for regular users. This paper presents PrefixWall, a system that secures multi-tenant LLM serving systems against APC side channels without sacrificing performance and efficiency. PrefixWall monitors cache reuse across users, flags suspicious sharing, and selectively isolates prefixes, restricting their reuse only when necessary. Evaluation shows that PrefixWall enables up to 70% higher cache reuse and 30% lower inference latency compared to existing defenses that isolate users. PrefixWall's lightweight design demonstrates how security in LLM serving does not have to come at the cost of unnecessarily reduced performance or unbearable overheads.

While Secure Aggregation (SA) protects update confidentiality in Cross-silo Federated Learning, it fails to guarantee aggregation integrity, allowing malicious servers to silently omit or tamper with updates. Existing verifiable aggregation schemes rely on heavyweight cryptography (e.g., ZKPs, HE), incurring computational costs that scale poorly with model size. In this paper, we propose a lightweight architecture that shifts from extrinsic cryptographic proofs to Intrinsic Proofs. We repurpose backdoor injection to embed verification signals directly into model parameters. By harnessing Catastrophic Forgetting, these signals are robust for immediate verification yet ephemeral, naturally decaying to preserve final model utility. We design a randomized, single-verifier auditing framework compatible with SA, ensuring client anonymity and preventing signal collision without trusted third parties. Experiments on SVHN, CIFAR-10, and CIFAR-100 demonstrate high detection probabilities against malicious servers. Notably, our approach achieves over $1000\times$ speedup on ResNet-18 compared to cryptographic baselines, effectively scaling to large models.

Machine learning backdoors have the property that the machine learning model should work as expected on normal inputs, but when the input contains a specific $\textit{trigger}$, it behaves as the attacker desires. Detecting such triggers has been proven to be extremely difficult. In this paper, we present a novel and explainable approach to detect and eliminate such backdoor triggers based on active paths found in neural networks. We present promising experimental evidence of our approach, which involves injecting backdoors into a machine learning model used for intrusion detection.

We introduce a novel copy-protection method for industrial control software. With our method, a program executes correctly only on its target hardware and behaves differently on other machines. The hardware-software binding is based on Physically Unclonable Functions (PUFs). We use symbolic execution to guarantee the preservation of safety properties if the software is executed on a different machine, or if there is a problem with the PUF response. Moreover, we show that the protection method is also secure against reverse engineering.

Instruction hierarchy (IH) defines how LLMs prioritize system, developer, user, and tool instructions under conflict, providing a concrete, trust-ordered policy for resolving instruction conflicts. IH is key to defending against jailbreaks, system prompt extractions, and agentic prompt injections. However, robust IH behavior is difficult to train: IH failures can be confounded with instruction-following failures, conflicts can be nuanced, and models can learn shortcuts such as overrefusing. We introduce IH-Challenge, a reinforcement learning training dataset, to address these difficulties. Fine-tuning GPT-5-Mini on IH-Challenge with online adversarial example generation improves IH robustness by +10.0% on average across 16 in-distribution, out-of-distribution, and human red-teaming benchmarks (84.1% to 94.1%), reduces unsafe behavior from 6.6% to 0.7% while improving helpfulness on general safety evaluations, and saturates an internal static agentic prompt injection evaluation, with minimal capability regression. We release the IH-Challenge dataset (https://huggingface.co/datasets/openai/ih-challenge) to support future research on robust instruction hierarchy.

Generative AI systems increasingly expose powerful reasoning and image refinement capabilities through user-facing chatbot interfaces. In this work, we show that the naïve exposure of such capabilities fundamentally undermines modern deepfake detectors. Rather than proposing a new image manipulation technique, we study a realistic and already-deployed usage scenario in which an adversary uses only benign, policy-compliant prompts and commercial generative AI systems. We demonstrate that state-of-the-art deepfake detection methods fail under semantic-preserving image refinement. Specifically, we show that generative AI systems articulate explicit authenticity criteria and inadvertently externalize them through unrestricted reasoning, enabling their direct reuse as refinement objectives. As a result, refined images simultaneously evade detection, preserve identity as verified by commercial face recognition APIs, and exhibit substantially higher perceptual quality. Importantly, we find that widely accessible commercial chatbot services pose a significantly greater security risk than open-source models, as their superior realism, semantic controllability, and low-barrier interfaces enable effective evasion by non-expert users. Our findings reveal a structural mismatch between the threat models assumed by current detection frameworks and the actual capabilities of real-world generative AI. While detection baselines are largely shaped by prior benchmarks, deployed systems expose unrestricted authenticity reasoning and refinement despite stringent safety controls in other domains.

Adversarial examples can represent a serious threat to machine learning (ML) algorithms. If used to manipulate the behaviour of ML-based Network Intrusion Detection Systems (NIDS), they can jeopardize network security. In this work, we aim to mitigate such risks by increasing the robustness of NIDS towards adversarial attacks. To that end, we explore two adversarial methods for generating malicious network traffic. The first method is based on Generative Adversarial Networks (GAN) and the second one is the Fast Gradient Sign Method (FGSM). The adversarial examples generated by these methods are then used to evaluate a novel multilayer defense mechanism, specifically designed to mitigate the vulnerability of ML-based NIDS. Our solution consists of one layer of stacking classifiers and a second layer based on an autoencoder. If the incoming network data are classified as benign by the first layer, the second layer is activated to ensure that the decision made by the stacking classifier is correct. We also incorporated adversarial training to further improve the robustness of our solution. Experiments on two datasets, namely UNSW-NB15 and NSL-KDD, demonstrate that the proposed approach increases resilience to adversarial attacks.

Spoofing attacks are among the most destructive cyber threats to terrestrial systems, and they become even more dangerous in space, where satellites cannot be easily serviced, and operators depend on accurate telemetry to ensure mission success. When telemetry is compromised, entire spaceborne missions are placed at risk. Prior work on spoofing has largely focused on attacks from Earth, such as injecting falsified uplinks or overpowering downlinks with stronger radios. In contrast, onboard spoofing originating from within the satellite itself remains an underexplored and underanalyzed threat. This vector is particularly concerning given that modern satellites, especially small satellites, rely on modular architectures and globalized supply chains that reduce cost and accelerate development but also introduce hidden risks. This paper presents an end-to-end demonstration of an internal satellite spoofing attack delivered through a compromised vendor-supplied component implemented in NASA's NOS3 simulation environment. Our rogue Core Flight Software application passed integration and generated packets in the correct format and cadence that the COSMOS ground station accepted as legitimate. By undermining both onboard estimators and ground operator views, the attack directly threatens mission integrity and availability, as corrupted telemetry can bias navigation, conceal subsystem failures, and mislead operators into executing harmful maneuvers. These results expose component-level telemetry spoofing as an overlooked supply-chain vector distinct from jamming or external signal injection. We conclude by discussing practical countermeasures-including authenticated telemetry, component attestation, provenance tracking, and lightweight runtime monitoring-and highlight the trade-offs required to secure resource-constrained small satellites.

Code agents powered by large language models can execute shell commands on behalf of users, introducing severe security vulnerabilities. This paper presents a two-phase security analysis of the OpenClaw platform. As an open-source AI agent framework that operates locally, OpenClaw can be integrated with various commercial large language models. Because its native architecture lacks built-in security constraints, it serves as an ideal subject for evaluating baseline agent vulnerabilities. First, we systematically evaluate OpenClaw's native resilience against malicious instructions. By testing 47 adversarial scenarios across six major attack categories derived from the MITRE ATLAS and ATT&CK frameworks, we have demonstrated that OpenClaw exhibits significant inherent security issues. It primarily relies on the security capabilities of the backend LLM and is highly susceptible to sandbox escape attacks, with an average defense rate of only 17%. To mitigate these critical security gaps, we propose and implement a novel Human-in-the-Loop (HITL) defense layer. We utilize a dual-mode testing framework to evaluate the system with and without our proposed intervention. Our findings show that the introduced HITL layer significantly hardens the system, successfully intercepting up to 8 severe attacks that completely bypassed OpenClaw's native defenses. By combining native capabilities with our HITL approach, the overall defense rate improves to a range of 19% to 92%. Our study not only exposes the intrinsic limitations of current code agents but also demonstrates the effectiveness of human-agent collaborative defense strategies.

As open-weights generative AI rapidly proliferates, the ability to synthesize hyper-realistic media has introduced profound challenges to digital trust. Automated disinformation and AI-generated imagery have made robust digital provenance a critical cybersecurity imperative. Currently, state-of-the-art invisible watermarks operate within one of two primary mathematical manifolds: the spatial domain (post-generation pixel embedding) or the latent domain (pre-generation frequency embedding). While existing literature frequently evaluates these models against isolated, classical distortions, there is a critical lack of rigorous, comparative benchmarking against modern generative AI editing tools. In this study, we empirically evaluate two leading representative paradigms, RivaGAN (Spatial) and Tree-Ring (Latent), utilizing an automated Attack Simulation Engine across 30 intensity intervals of geometric and generative perturbations. We formalize an "Adversarial Evasion Region" (AER) framework to measure cryptographic degradation against semantic visual retention (OpenCLIP > 75.0). Our statistical analysis ($n=100$ per interval, $MOE = \pm 3.92\%$) reveals that these domains possess mutually exclusive, mathematically orthogonal vulnerabilities. Spatial watermarks experience severe cryptographic degradation under algorithmic pixel-rewriting (exhibiting a 67.47% AER evasion rate under Img2Img translation), whereas latent watermarks exhibit profound fragility against geometric misalignment (yielding a 43.20% AER evasion rate under static cropping). By proving that single-domain watermarking is fundamentally insufficient against modern adversarial toolsets, this research exposes a systemic vulnerability in current digital provenance standards and establishes the foundational exigence for future multi-domain cryptographic architectures.

This paper proposes PRoADS, a provably secure and robust audio steganographic framework based on audio diffusion models. As a generative steganography scheme, PRoADS embeds secret messages into the initial noise of diffusion models via orthogonal matrix projection. To address the reconstruction errors in diffusion inversion that cause high bit error rates (BER), we introduce Latent Optimization and Backward Euler Inversion to minimize the latent reconstruction and diffusion inversion errors. Comprehensive experiments demonstrate that our scheme sustains a remarkably low BER of 0.15% under 64 kbps MP3 compression, significantly outperforming existing methods and exhibiting strong robustness.

Deceptive patterns, dark patterns, and manipulative user interfaces (UI) are a widely used design strategy that manipulates users to act against their own interests in pursuit of shareholder aims. These patterns may particularly affect people with less education, visual impairments, and older adults. Yet, access is a critical feature of the user experience (UX), development standards, and law. We considered whether and how the Web Content Accessibility Guidelines (WCAG) and related legislation, like the European Accessibility Act (EAA), could act as a tool against deceptive patterns. We used heuristic evaluation to analyze whether and how deceptive patterns violate or conform to these guidelines and legal statutes. Although statistical analysis revealed no significant differences by pattern type, we identified three patterns implicated by the WCAG guidelines: Countdown Timer, Auto-Play, and Hidden Information. We offer this approach as one tool in the fight against UI-based deception and in support of inclusive design.

Retrieval-Augmented Generation (RAG) systems extend large language models (LLMs) with external knowledge sources but introduce new attack surfaces through the retrieval pipeline. In particular, adversaries can poison retrieval corpora so that malicious documents are preferentially retrieved at inference time, enabling targeted manipulation of model outputs. We study gradient-guided corpus poisoning attacks against modern RAG pipelines and evaluate retrieval-layer defenses that require no modification to the underlying LLM. We implement dual-document poisoning attacks consisting of a sleeper document and a trigger document optimized using Greedy Coordinate Gradient (GCG). In a large-scale evaluation on the Security Stack Exchange corpus (67,941 documents) with 50 attack attempts, gradient-guided poisoning achieves a 38.0 percent co-retrieval rate under pure vector retrieval. We show that a simple architectural modification, hybrid retrieval combining BM25 and vector similarity, substantially mitigates this attack. Across all 50 attacks, hybrid retrieval reduces gradient-guided attack success from 38 percent to 0 percent without modifying the model or retraining the retriever. When attackers jointly optimize payloads for both sparse and dense retrieval signals, hybrid retrieval can be partially circumvented, achieving 20-44 percent success, but still significantly raises attack difficulty relative to vector-only retrieval. Evaluation across five LLM families (GPT-5.3, GPT-4o, Claude Sonnet 4.6, Llama 4, and GPT-4o-mini) shows attack success ranging from 46.7 percent to 93.3 percent. Cross-corpus evaluation on the FEVER Wikipedia dataset (25 attacks) yields 0 percent attack success across all retrieval configurations.

Embedded cryptography stands or falls on entropy quality, yet small devices have few trustworthy sources and little tolerance for heavyweight protocols. We build a Quantum Entropy as a Service (QEaaS) system that moves QRNG-derived entropy from a Quantis device to ESP32-class clients over post-quantum-secured channels. On the server side, the design exposes two paths: direct quantum entropy through a custom OpenSSL provider and mixed entropy through the Linux system pool. On the client side, we extend libcoap's Zephyr support, integrate wolfSSL-based DTLS 1.3 into the CoAP stack, and add a BLAKE2s entropy pool that preserves the standard Zephyr extraction interface while introducing an injection API for server-provided entropy. Benchmarks on ESP32 hardware, targeting 100 iterations per configuration, show that ML-KEM-512 completes a DTLS 1.3 handshake in 313 ms on average without certificate verification, 35% faster than ECDHE P-256. Pairing ML-KEM-512 with ML-DSA-44 lowers the mean to 225 ms. Certificate verification adds roughly 194 ms for ECDSA but only 17 ms for ML-DSA-44, so the fully post-quantum configuration remains 63% faster than classical ECDHE P-256 with ECDSA even under full verification. Local BLAKE2s pool operations stay below 0.1 ms combined. On this platform, post-quantum key exchange and authentication are not only feasible; they are faster than the classical baseline.