Papers

Large Language Models are increasingly deployed in Security Operations Centers for log analysis tasks including summarization, alert triage, and threat investigation. These systems ingest logs from external-facing services and process network logs as natural language contexts to generate security insights. We demonstrate that this architectural pattern introduces a critical vulnerability: adversaries can embed prompt injection payloads in log-generating fields that persist in storage and are executed when analysts query the LLM, achieving what we term passive prompt injection. We present LogInject, a systematic framework for evaluating these threats. Using LogInject-1.0, a benchmark of 12,847 log entries including 2,569 adversarial samples, we evaluate three production LLMs across four attack objectives: activity concealment, false positive generation, information exfiltration, and output hijacking. Our findings reveal an up to 88.2% attack success rate (83.4% average across models) under the baseline conditions. We introduce Context Stitching, a novel technique that fragments payloads across multiple log entries to evade stateless filters while exploiting LLM long-context reasoning, achieving a 76.4% success rate. As mitigation, we evaluate layered defenses by combining input filtering, prompt hardening, and output validation, demonstrating a 90.4% attack reduction, although 8.4% residual vulnerability persists. Our results establish that LLM-based log analysis creates an inherent confused deputy vulnerability where untrusted data and trusted instructions compete indistinguishably for model attention, requiring defense in-depth architectures and continued human oversight for security-critical decisions.

With the rapid growth of mobile applications, user data privacy has become an increasing concern. While privacy policies describe how apps collect and share data, platforms such as Google Play provide Data Safety labels intended to summarize these practices. Because these disclosure channels are declared separately, they may present inconsistent representations of app data practices, creating uncertainty for users and regulators. In this work, we conducted a large-scale empirical study of disclosure consistency across 6,051 Android apps. Using an LLM-based extraction framework and a unified schema over 14 Google Play data categories and two operations (collection and sharing), we measure per-app and per-category consistency and introduce a sensitivity-weighted risk score that emphasizes high-risk data types. We find that misalignment disproportionately affects sensitive categories such as personal information and device identifiers, with sharing disclosures exhibiting lower consistency than collection disclosures. Elevated privac risk is concentrated in app categories associated with persistent monitoring and communication. Overall, our findings highlight structural gaps in current disclosure mechanisms and underscore the need for stronger verification and greater transparency in platform-level privacy reporting.

Sandboxing remains a core technique for observing suspicious program behavior, yet environment-aware malware increasingly suppresses execution when analysis is suspected. Prior generations of sandbox evasion focused on virtualization artifacts, timing discrepancies, and wear-and-tear realism. In this paper, we present the first systematic measurement study of AI-environment artifacts as a new sandbox-evasion surface. We operationalize this realism gap through AIprint, a probe framework that captures persistent artifacts left behind by AI-capable software ecosystems, including AI-assistant configuration directories, model caches, environment variables, local inference services, and package dependencies. We systematically extract 450 unique artifacts from 284 open-source AI projects on GitHub, compile them into unprivileged Windows probes, and evaluate them across seven commercial and open-source sandbox backends together with three AI-capable reference hosts. Our results show that traditional VM-detection baselines fail to reliably distinguish real AI-capable systems from modern sandboxes, whereas twelve AI-environment artifacts appear on the reference hosts and on none of the evaluated backends. A controlled 214-step installation experiment establishes a causal relationship between AI tool and package installation and measurable AI-environment artifact accumulation, while adaptive spoofing experiments reveal a fundamental operational asymmetry: reproducing convincing AI software environments is substantially more expensive than detecting shallow spoofing.

Dependency resolution at a specified point in time in the past can provide insight into software evolution in software ecosystems and facilitate the design of dynamic metrics (e.g., dependency freshness, dependency update rhythm). However, dependency resolution at specified points in time is not possible in major software ecosystems due to a lack of support from package management tools. The goal of this paper is to aid practitioners and researchers in analyzing the state of the ecosystem dependency graph at release points using an enriched dataset with dependency resolution at release points for npm, PyPI, and crates.io packages. In this paper, we present a methodology to construct dependency resolution at release points of packages in software ecosystems, which we enrich with vulnerability data from OSV. We apply our methodology to construct CHRONO-RESOLUTION, a dataset of dependency resolution at release points for npm, PyPI, and crates.io packages, and release it for future research.

Pujol and Desfontaines asked whether a private histogram can allow more error on larger counts and use that slack to protect members of larger groups more strongly. We study this question for fixed disjoint groups under add-or-remove-one adjacency. The privacy budget $v(n)$ depends on the affected count, is nonincreasing, and must bound both Rényi-divergence directions at every order. This is the count-dependent form of zero-concentrated differential privacy (zCDP) studied here. The original strict relative-error condition is impossible at count zero. We therefore make the boundary tolerance explicit by requiring $\mathbb{E}\lvert\widehat{x}_i-x_i\rvert < r\max\{x_i,1\}$, without changing the requirement at any positive count. Our main result determines the best dependence on group size. For the upper bound, we directly specialize an existing shifted-transformation framework. The resulting shifted-log Gaussian mechanism has a certified budget $v(n)=O_r(n^{-2})$. Conversely, for every fixed $0<r<1$, any mechanism satisfying the same positive-count utility requirement and count-dependent zCDP must have $v(n)=\Omega_r(n^{-2})$. Thus the inverse-square rate is optimal under the repaired formulation. A many-count information argument further places the leading coefficient in the large-count-then-small-error limit between $\pi/(4e^2)$ and $1/\pi$, a factor below three. At $r=1$, a data-independent release meets the repaired criterion with zero privacy loss.

The Internet of Things (IoT) was introduced almost two decades ago. In the past two decades, technology has seen huge advancements. Many devices have become powerful and have less power consumption. Many IoT architectures and environments were introduced to help make life easier, especially in wearable devices. The market for these wearable devices has constantly increased over the years and is expected to reach its maximum in the next couple of years. They also pose a threat to users' privacy and security because they constantly store and transmit personal information such as location, heart rate, and other sensitive data. Therefore, addressing the security vulnerabilities is a crucial aspect of this research. This paper presents a hardware-assisted, energy-efficient, low-overhead security solution for wearable devices. Specifically, two Physical Unclonable Function (PUF) architectures: Arbiter PUF and Hybrid Oscillator Arbiter (HOA) PUF are analyzed for integration in IoT systems. The result shows that Arbiter PUF consumes 25 $\mu$W, whereas HOA PUF consumes only 2.7 $\mu$W to generate keys for cryptographic purposes. These architectures introduce minimal power overhead while providing robust security, making them well suited for resource-constrained IoT ecosystems.

People use language models for practical questions whose answers are difficult to verify. We show that models exhibit covert value leakage: the information they provide is influenced by their own values, without this influence being disclosed to the user. In one of our evaluations, the user is considering investing in an AI company and wants to know how likely the AI bubble is to pop. Claude Opus 4.8 gives a lower probability when the company under consideration is Anthropic rather than OpenAI. Yet Claude mostly fails to disclose this influence to the user. Covert value leakage is a form of misalignment because it goes against the user's preferences and is likely to mislead them. To investigate this phenomenon, we introduce a suite of evaluations to quantify value leakage and whether models disclose it. We find that models are influenced by different types of values, including preferences for morally good outcomes, for the company that developed them, and for some human leisure activities over others. We often observe large differences among frontier models on the same evaluation. For example, on a Fermi-estimation task, Claude models falsely claim to give unbiased answers in their chain-of-thought, while Qwen models explain how their values bias their answers. Value leakage is a failure mode distinct from sycophancy and reward hacking, and current alignment training and evaluations do not adequately address it.

AI coding agents produce code faster than humans can review it. In our approach, the prover is the judge of whether the code is correct. Under a verifier-driven loop, AI agents wrote and verified bare-metal security software in Ada/SPARK spanning classical and post-quantum cryptography, TLS 1.3, IKEv2, X.509, and a Matrix client. GNATprove discharged 49,280 proof obligations, established functional correctness for selected primitives, and proved the absence of run-time errors for the rest, at roughly 20-40 times lower supervision cost than comparable hand verification. GNATprove alone was insufficient: some defects could not be detected and were resolved using known-answer tests, interoperability, or human review of specifications. Given weak checks, the agent tried to bypass them and reported success. We report where each layer caught faults and draw the central lesson: what an agent can be trusted to establish is bounded by the strength of its feedback.

Federated learning (FL) enables multi-institutional training on clinical text without sharing raw data, but gradient inversion can reconstruct sensitive information from shared model updates. The extent of this leakage for radiology reports, and the role of tokenizer design, remains unclear. We quantify gradient-based text reconstruction in FL and compare privacy risk across three tokenizers with the model architecture held fixed. Six FL clients trained a GPT-2-style transformer (sequence length 32) on public radiology corpora (368,751 diagnostic reports, 98,206 discharge summaries, 1,500 MIMIC-CXR free-text reports) using the GPT-2, RadBERT, and LLaMA-2 tokenizers at batch sizes of 64, 128, and 256. Assuming an active malicious server that modifies the shared architecture before distribution, we applied analytic gradient inversion and measured reconstruction fidelity over five runs. Exact sentence reconstruction ranged from 31% to 44% across tokenizers (30.6-43.5% across the 27 tokenizer x dataset x batch-size cells). At batch size 64 on the Discharge dataset, accuracy was 42.1% (GPT-2), 42.3% (RadBERT), and 39.4% (LLaMA-2), decreasing to 37.3%, 37.2%, and 34.3% at batch size 256. S-BLEU declined as batch size grew (GPT-2: 0.44 to 0.33; RadBERT: 0.48 to 0.35). RadBERT yielded the highest reconstruction fidelity and recovered the most clinical terms (18.1% of a 1,440-term reference vocabulary, vs 12.5% for GPT-2 and 9.4% for LLaMA-2), yet no tokenizer prevented leakage. Substantial portions of report text are therefore recoverable from FL gradients even at larger batch sizes and with domain-specific tokenizers. Tokenizer design influences leakage severity and is a privacy-relevant decision, not only a utility one; safeguards such as secure aggregation and differential privacy are likely necessary to meet HIPAA and GDPR requirements for FL in radiology NLP.

Production LLM-agent frameworks ship control primitives – human-in-the-loop approval gates, run cancellation, and execution timeouts – whose names and documentation imply barrier semantics: while a run is paused, cancelled, or timed out, no gated side effect executes. This contract holds on none of six widely used open-source frameworks. Model-free differential probes isolate a recurring sibling leak – an approval gate suspends its own branch while a sibling's effect executes during the pause, defeating rejection – in every framework shipping a pre-execution gate (five of six, four execution models, two language runtimes), and confirm replay double-execution, cancellation orphans, and timeout zombies. The hazard is reachable: frontier models emit the leak-triggering plan shape at rates up to 14%, and live models driving unmodified frameworks leak 215 of 1,200 runs (P(leak | emitted)=1.00); on naturalistic tau-bench episodes models serialize writes – the everyday gap is latent – while injection induces it deterministically and a 13-incident public corpus corroborates the replay and cancellation failures. We repair the gaps with SOUNDGATE, an environment-external Rust gate through which every side effect must be admitted, enforcing hold-until-decided, reject-cancels, dedup-on-replay, and fence-on-cancel under a stated complete-mediation contract, discharged for network egress by two kernel-enforced routes. The admission core is mechanically verified (Verus; TLA+/TLC to 7.5e7 states; TLAPS; Loom on the deployed Rust) and bridged to code by differential conformance over 1.2e7 operations with zero divergences. Under that contract SOUNDGATE blocks every measured violation on all six frameworks while releasing legitimate effects: gated tau-bench episodes complete with zero refusals at ~1 ms per write, and durable admission sustains ~12k admissions per second.

Penetration testing traditionally evaluates whether adversaries can exploit weaknesses in software, infrastructure, configurations, or operational controls to achieve security-relevant compromise. This paradigm remains necessary for AI-enabled systems, but it is no longer sufficient. In such systems, adversaries may influence prompts, retrieved content, sensor inputs, training data, memory, tools, or human-AI interaction loops to alter system behavior without directly compromising the underlying infrastructure. This paper reframes penetration testing for AI-enabled systems as objective-driven behavioral evaluation. We define an AI-enabled system as one in which learned models materially influence behavior affecting operational outcomes, and we define AI-enabled penetration as the feasible induction of AI-governed behavior that violates one or more operational objectives under an explicit threat model. This definition preserves conventional penetration testing while extending it to adversarial pathways such as prompt injection, indirect prompt injection, data poisoning, sensor manipulation, retrieval poisoning, tool misuse, and agentic misalignment. We further propose a testing workflow that identifies operational objectives, maps AI-governed behavior, analyzes adversarial influence surfaces, defines behavioral failure criteria, executes scenario-based tests, and reports evidence linking adversarial action to objective violation. A running example involving an AI-enabled security operations center assistant illustrates how penetration may occur through behavioral influence rather than infrastructure compromise. Together, the definitions, workflow, and example provide a technical framework for evaluating adversarial success in deployed AI-enabled systems.

Reusable skills are becoming a fundamental building block of Large Language Model (LLM) agents, enabling capabilities to be packaged, shared, and reused across diverse applications. However, existing security research primarily focuses on prompt injection and runtime execution, leaving security risks throughout the broader skill lifecycle largely unexplored. In this paper, we present SkillSec-Eval, a lifecycle-aware framework for systematically evaluating the security of reusable agent skills. We first characterize the skill lifecycle and develop a threat taxonomy spanning repository admission, semantic retrieval, planner selection, execution, and skill evolution. We then instantiate this taxonomy in SkillSec-Eval and conduct a comprehensive empirical evaluation using a repository of 327 real-world skills. Our study demonstrates that vulnerabilities arise at multiple lifecycle stages beyond execution, highlighting the need for lifecycle-aware security analysis of reusable agent skills.

Open source software is vulnerable to supply-chain attacks through transitive dependencies, especially malicious code injected into NPM packages. Existing detectors often inadequately model obfuscated behavior, overlook JavaScript's object-centric features, poorly coordinate static and dynamic analysis, and lose semantic information during behavior abstraction. We propose ProfMalPlus, a malicious NPM package detector combining object-sensitive behavior graphs with coordinated LLM reasoning over annotated code slices. It identifies installation commands and entry files, then constructs graphs capturing sensitive APIs, third-party calls, and unresolved calls. From these graphs, ProfMalPlus extracts security-relevant slices and adds inline static analysis evidence. Local judge agents independently assess each slice. Self-consistency consolidates repeated judgements to reduce LLM variance, while a global judge synthesizes their reports into an entry-level verdict. For undetermined cases, a router selects either third-party enrichment, which adds registry derived module and method semantics, or dynamic augmentation, which executes the package in a sandbox to resolve runtime dependent behavior. The enriched evidence is fed back for reassessment. Finally, a localization agent reports malicious code snippets with explanations. ProfMalPlus achieves a 98.1% F1-score, outperforming state-of-the-art detectors by 3.5% to 52.6%. It also identified 597 previously unknown malicious packages, all confirmed and removed from NPM.

This work presents a compact RISC-V extension for Post-Quantum Cryptography (PQC) called HORCRUX, which provides a unified Instruction-Set Extension (ISE) supporting all NIST-approved PQC algorithms. HORCRUX addresses the difficult trade-off between crypto-agility, high performance, and low resource consumption in constrained environments, a balance typically missing in hardware extensions that focus on limited PQC subsets. By targeting shared kernels across ML-KEM, MLDSA, SLH-DSA, HQC, and Falcon, the extension introduces new RISC-V instructions executed by a resource-efficient, tightly coupled coprocessor. This architecture is specifically optimized for embedded systems with strict energy budgets and limited area. Experimental evaluation on a Zynq UltraScale+ FPGA demonstrates speedups of up to 129x for hash-based, 9x for lattice-based, and 27x for code-based schemes, while adding fewer than 21k LUTs and 4.4k FFs. ASIC results from postsynthesis characterization in 65 nm CMOS are also reported, alongside a rigorous power characterization to validate the architecture's energy efficiency. The extension's modular structure maintains backward compatibility with standard RISC-V cores, offering a scalable solution for deploying PQC on constrained embedded systems.

Whistleblowers are a key safeguard against organizational wrongdoing, but the threat of retaliation deters reporting. Existing whistleblower-protection proposals lack formal privacy guarantees, and existing differential privacy mechanisms do not directly target the natural threat model – one in which the audited organization itself observes auditor selection decisions and uses them to identify reporters. We formalize protection against a strong-adversary threat model as per-report $(0, \delta)$-differential privacy on the transcript of audit selections. Within this framework we prove that a natural approach – randomized response applied at the selection step – can never outperform uniform random auditing by more than $\delta$ at any horizon. We then give a generic mechanism that reduces private auditing to private continual counting: any $(0, \delta)$-DP continual counter plugs in by post-processing, and the audit transcript inherits the same per-report guarantee. Instantiating the reduction with a recent work in continual counting yields per-report $(0, \delta)$-DP with noise scaling as $O(\sqrt{\log T})$ across a horizon of $T$ audit decisions. A utility theorem shows that the selection error vanishes whenever the noisy report gap between the most-reported organization and the runner-up grows faster than $\sqrt{\log T}$. Simulations show a substantial improvement over randomized response.

Biometric template protection (BTP) must secure stored templates while tolerating intra-class variations. Existing methods rely on protected-domain similarity matching, error correction, or predefined-template mappings, potentially retaining exploitable similarity structures, introducing helper-data risks, depending on artificial targets, or coupling protection to specific modalities. Storing only cryptographic hash digests eliminates directly comparable representations and conceals pre-hash templates, but hash-based exact-match verification requires genuine samples to generate identical intermediate templates before hashing. Identity classification is naturally suited to this requirement because it maps variable biometric samples to stable and discriminative identity-level outputs. Based on this insight, we propose Multiple Permuted-Label Classifier Encoding (MPLCE). Through classifier-specific label permutations, MPLCE assigns each identity different labels across multiple classifiers. The predicted labels are encoded and concatenated to form an intermediate template, preventing repeated encodings of a single identity label and enlarging the effective candidate space while preserving classification consistency. The template is randomized with an application-specific XOR string and cryptographically hashed, enabling exact-match verification without error correction codes or biometric-dependent helper data. Using modality-specific classifiers, MPLCE retains the same template generation and protection procedure across modalities. On four face and two iris datasets, MPLCE achieves competitive performance, including a GAR of 98.61% at a FAR of 5.51\(\times\)10\textsuperscript{-5}% on YTF and a GAR of 99.10% at a FAR of 0.00% on CASIA-Iris-Lamp. Security analyses and attack evaluations support its irreversibility, revocability, and unlinkability under the threat model.

Encrypted mempools are designed to hide transaction contents until execution order is fixed, preventing many victim dependent forms of maximal extractable value. This paper studies a different class of attack in the form of self-authored state manipulation, in which the attacker knows its own transaction and owns a downstream claim on the state that transaction changes. Perpetual futures funding is a canonical example. The funding signal determines a transfer rate, while receiving side open interest is the transfer base. In a commit then reveal mempool, an adaptive corrective transaction cannot enter the already committed batch. Privacy can therefore create an economic reaction gap even when cryptographic decryption overhead is negligible. We microfound correction through executable arbitrage opportunities. Correctors choose order size against local price impact and inventory cost, while the protocol information schedule determines which opportunities are actionable. The ordering barrier removes ordinary adaptive searchers from the closed stage. It therefore yields a closed stage correction rate below the adaptive correction rate whenever positive adaptive capacity becomes available after reveal. The distortion entering a funding window is multiplied by an explicit response factor. Transaction privacy can also reduce capitalization of predictable funding into entry prices, producing a second amplification channel. The resulting local security index separates attacker blindness, correction shielding, and capitalization shielding.

Android memory forensics recovers evidence that never touches disk: decrypted messages, session credentials, and the live internal state of a running application. The tools that perform this recovery depend on debug symbols embedded in libart.so, the Android Runtime library, to locate data structures and interpret their layout. Across recent releases Google has stripped most of that information from the binaries that ship on consumer phones as part of a broader security hardening effort, yet no prior work has measured how far the stripping has progressed on the devices examiners actually encounter, or whether the memory architecture beneath the stripped surface still resembles what the forensics literature describes. This paper measures the gap between the unstripped development builds researchers use and the production builds that ship on real hardware, using binaries extracted from Pixel factory images across Android 8 through Android 15. Static symbols fell from 20,495 entries to zero, dynamic symbols dropped by roughly 60 percent, and source file references disappeared entirely. A compressed fallback section in the Android 15 binary restores thousands of function names but carries no structure layouts. Source code review and memory map comparisons across Android 8 and 15 show that the heap spaces, garbage collector infrastructure, and allocation bitmaps remain structurally intact, with visible changes limited to naming and arithmetic optimization. Live validation on a rooted, fully stripped Pixel 7 confirms that the runtime entry point is still locatable through the dynamic symbol table, and that structural offsets pulled from a version-matched development build resolve to valid pointers inside production memory.

The Elliptic Curve Discrete Logarithm Problem (ECDLP) is a fundamental problem in cryptography, and reducing the resource requirements of quantum algorithms for solving ECDLP is an important goal. In this work, we present a space-efficient quantum algorithm for solving the ECDLP over prime fields, achieving an implementation with only $3n+6\lfloor \log_2 n \rfloor+O(1)$ logical qubits and $1056n^3/\log_2 n+O(n^2)$ Toffoli gates, where $n$ is the bit-length of the prime. For a 256-bit prime-field curve, our construction requires only 835 logical qubits, reducing the previous best estimates of 1098 and 1175 logical qubits by Chevignard et al. [EUROCRYPT 2026] and Babbush et al. [ArXiv Preprint 2026], respectively. The key to our improvement is a new space-efficient reversible modular inversion circuit, which addresses the dominant space bottleneck in affine-coordinate point addition. Starting from the extended Euclidean algorithm (EEA), we refine the register-sharing technique of Proos and Zalka by introducing length registers and location-controlled arithmetic to compactly store and update intermediate variables. We further optimize the reversible update procedures and construct the corresponding controlled arithmetic circuits, resulting in a modular inversion circuit implemented by only $2n+6\lfloor \log_2 n \rfloor+O(1)$ logical qubits and $229n^2+O(n\log_2 n)$ Toffoli gates. This modular inversion circuit together with mid-circuit measurements and classical feed-forward operations provides a space-efficient controlled affine point-addition circuit and a complete implementation of Shor's algorithm for ECDLP.

Large language model (LLM)-based intrusion detection systems (IDS) are increasingly studied for security monitoring, yet their robustness against feasible traffic manipulation remains largely empirical. We present Traffic-Aware Randomized Smoothing (TA-RS), a classifier-agnostic certified defense that injects Gaussian noise exclusively into the directly controllable (DC) subspace – features a remote attacker can modify – during both fine-tuning and certification, aligning the smoothing distribution with the attacker-controllable subspace. We identify a critical prerequisite: applying standard randomized smoothing to clean-trained LLM-IDS yields weak certified accuracy in three of four (model, dataset) pairs tested (14-33%, at or below random) and only 57% in the fourth (43 pp below the noise-augmented result); noise-augmented fine-tuning recovers to 68-100% on two of three benchmark datasets (at sigma=0.25). At the L_inf-equivalent threshold R_inf = epsilon*sqrt(|DC|) (epsilon=0.05), TA-RS achieves 55-100% certified accuracy on CIC-IDS-2018 and HIKARI-2021, with median certified radii (R approx 0.45-0.96) exceeding R_inf by 1.8-5x (across sigma=0.25-1.00). Against a fairly trained iso-trained RS baseline the residual advantage is dataset-dependent (4-19 pp on CIC-IDS-2018). The larger gap – up to 72 pp against an isotropic RS baseline that shares the DC-noise-augmented training recipe – primarily reflects the training-certification mismatch rather than DC alignment alone: isotropic test-time noise perturbs uncontrollable features the attacker cannot exploit, triggering abstention rates up to 68%. RT-IoT2022 probes the limits of the method: it fails under the default fine-tuning recipe but recovers to 76%/69% certified accuracy (LLaMA3-8B/Qwen3-8B) when noise augmentation is increased.

Safeguarding patient privacy while enabling meaningful healthcare data use remains critical under GDPR and HIPAA. Existing compliance methods are manual, error-prone, and separate policy audits from data-level assessments. This paper presents PriEval-Protect, a two-phase framework for unified privacy risk evaluation and mitigation. The evaluation phase combines regulatory compliance scoring using a fine-tuned legal LLM with RAG, and technical analysis via encryption type, data architecture, and metrics including similarity, uncertainty, adversary success, and information gain/loss. A composite risk score uses weighted aggregation via Analytic Hierarchy Process. The protection phase recommends countermeasures including federated learning and differential privacy based on assessed risk. Results on hospital documents and datasets demonstrate regulation-aligned, explainable assessments, bridging legal conformance and data-level risk analysis.

The potential capabilities of quantum computers motivated the development of cryptographic protocols suitable for securing communication against adversaries with access to large fault-tolerant quantum computers. However, even though current quantum computers are limited in terms of size and precision, they can still be useful for finding loopholes and weaknesses in the post-quantum cryptographic protocols. In this work, we present an attempt to utilize the capabilities of Quantum Generative Adversarial Networks (QGANs), one of the promising architectures used in quantum machine learning, for this purpose. We describe an example application of QGAN architecture for the purpose of loading the probability distribution of the hash-based digital signatures into the memory of a quantum computer. Our results confirm that near-term hybrid quantum-classical methods possess capabilities required for this purpose. The presented approach can be used as a first step in the workflow, enabling the utilization of quantum computing for attacking post-quantum cryptographic primitives.

As AI agents grow in prevalence, so too do their risks to users, creating a need for robust and usable systems for agent permissions. Many recent proposals implement product-level permissions, where the same security policy is applied for all users of a given agent framework. Yet different users have varying needs, necessitating support for user-level permission policies in agentic AI systems. We survey 21 proposed user-level agent permissions systems and construct a taxonomy of approaches to specifying, deriving, and enforcing user-level permissions. We further analyze and compare against five commercial agents. We find that despite a wide diversity of proposals, sizable gaps remain in usability, transparency, and user control, particularly between academic and commercial agents.

Heterogeneous CPU-GPU workloads are increasingly used in safety-critical embedded systems, yet no existing approach provides joint attestation of their execution. Prior Control-Flow Attestation (CFA) techniques focus on CPU-side CFA, while GPU attestation is limited to static, load-time verification and does not provide runtime guarantees. As a result, runtime attacks on GPU kernels and violations of the CPU-GPU interaction contract remain unaddressed. We present WarpGuard, the first composite CFA framework for heterogeneous CPU-GPU workloads. WarpGuard verifies execution against a unified control-flow graph (CFG) that captures both CPU and GPU components. It extends prior CFA techniques in two ways: it enables runtime CFA of GPU kernels by tracing their execution against kernel-specific CFGs, and it monitors kernel launch events and enforces per-call site policies to detect violations at the CPU-GPU boundary. These extensions address challenges arising from GPU parallelism and cross-device interactions. We implement WarpGuard using software-based instrumentation, requiring no specialized hardware or binary modifications. Our evaluation on an NVIDIA Jetson Orin Nano shows that WarpGuard detects GPU-side control-flow and cross-boundary attacks. Across microbenchmarks, SPECAccel, and eight TensorRT inference workloads, WarpGuard incurs moderate overheads, suggesting practicality for embedded safety-critical settings.

When cast as the protector of a vulnerable user yet given no explicit capability boundary, a large language model (LLM) may respond not by acknowledging its limits but by claiming to have taken, or to be taking, a real-world protective action it cannot perform, such as contacting emergency services or administering care. We term this phenomenon Protective Capacity Hallucination (PCH): a self-referential misattribution in which a model, acting in a protective role, asserts physical or institutional agency exceeding its affordances as a language model. In a three-phase study spanning eight LLMs and 13,600 sessions, we find that PCH depends on both situational severity and interactional format. Across ordinary service domains, multi-party dialogic input drives PCH to near-ceiling levels in most models. In contrast, PCH remains at floor levels in all eight models when the same models are placed in intimate-partner conflict scenarios, despite the greater physical severity of those situations. We interpret PCH as the signature of a deployment-design gap between role assignment and capability-boundary specification: a by-product of partial alignment in which a universally trained pressure to help outruns a domain-selective specification of how to help. Because suppression tracks alignment coverage rather than severity, deployment-side specification of capability boundaries emerges as a general mitigation target.

We investigate which language model evasion attacks survive state-of-the-art adversarial fine-tuning, developing strategies that sweep the top 5 positions on the ELOQUENT 2026 Voight-Kampff leaderboard. While adversarial fine-tuning trivially closes the 2025 winning evasion recipes, we uncover a fundamental asymmetry in detector vulnerability: pushing generated text out of the detector's training distribution reliably defeats adversarial detection, whereas pulling it into the distribution (e.g., mimicking human training data) fails completely. Exploiting this, we introduce two novel out-of-distribution attack families - cross-decade register attacks and modernist stream-of-consciousness form. Both strategies easily bypass adversarial closure, achieving up to approximately 50x higher fool rates than previous methods while preserving naturalness. Furthermore, experiments show that the obvious deployer countermeasure (augmenting training data with period prose) fails to close the vulnerability. Our findings show that the tested detector families, including adversarially fine-tuned ones, exhibit persistent vulnerabilities under structural out-of-distribution shifts, a mechanism that directly powers our leading competition performance.

As quantum computing technology continues to mature, the US National Institute of Standards and Technology (NIST) has outlined a migration timeline for Post-Quantum Cryptography (PQC), recommending the deprecation of certain elliptic curve cryptography (ECC) by 2030. Furthermore, privacy-sensitive application scenarios, such as vehicular communications, require the use of anonymous certificates. However, existing anonymous certificate schemes are still largely based on ECC. Therefore, this study proposes a multivariate cryptography-based anonymous certificate scheme, aiming to design quantum-safe anonymous certificates suitable for privacy-sensitive application services. The proposed multivariate cryptography-based anonymous certificate scheme is supported by rigorous mathematical proofs and illustrated with computational cases.

To overcome data scarcity and privacy constraints in data collection, it has become standard practice across academia and industry to augment real training data with text-to-image (T2I)-generated synthetic data, a paradigm we term Real-Synthetic Mix-Training (RSMT). While substituting synthetic data for sensitive real samples is widely regarded as a means to mitigate privacy exposure of the substituted data, the risk to the remaining real samples that actively participate in training has remained largely unexamined. This work reveals, for the first time, that RSMT can substantially amplify privacy leakage of these real training samples. We establish a theoretical framework, RSMT Memorization Amplification, proving that incorporating synthetic data displaces real samples toward peripheral regions of the mixed feature space, in turn forcing the model to memorize them more aggressively. Guided by this foundation, we propose RSMixLeak to systematically assess this risk through membership inference attacks (MIAs). RSMixLeak comprises two variants depending on the adversary's capability. The non-adversarial variant audits a benign RSMT pipeline with an honest T2I provider, establishing a lower bound on the leakage induced by the intrinsic gap between real and T2I-generated data. The adversarial variant considers an adversary who controls the T2I model or contributes crafted data to the T2I provider, and deliberately enlarges this distributional gap on a target class via either high-level semantic attribute binding or imperceptible pixel-level coating, further amplifying leakage on real training data while improving downstream model utility. Motivated by these findings, we further propose a lightweight leakage propensity indicator computable from real data alone that reliably identifies high-risk datasets unsuitable for entering RSMT, as a self-assessable mitigation.

ColumnDisturb is a new data-disturbance error in which activations to an aggressor row cause bitflips in a victim row located hundreds of rows away (intra-subarray bitflips) and in victim rows in adjacent subarrays (inter-subarray bitflips). Intra-subarray ColumnDisturb can be tolerated by solutions (such as SALT and REGA) that operate at subarray granularity. However, to tolerate inter-subarray ColumnDisturb, such solutions must be extended with ColumnDisturb Protection (CDP), which performs additional Adjacent-Counter Increment (ACI) for the neighboring subarrays. The ACIs ensure that adjacent subarrays also undergo mitigation, even if they receive no demand activations. Unfortunately, because ACIs occur at a 200% rate relative to demand activations, they effectively increase the activations perceived by the bank to 3x, which causes significant slowdowns (17% at a TRHD of 500) and refresh overheads. The goal of our paper is to tolerate ColumnDisturb while incurring negligible overheads. We propose CODA, a ColumnDisturb mitigation that significantly reduces the rate of ACI required to securely tolerate ColumnDisturb. We present three variants of CODA. First, CODA-E (Evade), which leverages the insight that ACI can be skipped if the neighboring subarray receives a demand activation, and reduces ACI by 2x. Second, CODA-F (Fraction), which uses the timing duration of ColumnDisturb to do only a fractional increment for ACI, thereby reducing the rate of ACI by 2x-16x. Finally, CODA-G (Gangskip), which operates at multi-subarray granularity and skips ACI for neighboring subarrays within the same gang, further reduces overall ACI by 2x-8x. Overall, CODA reduces ACI by 12x-1300x, thereby making it possible to tolerate ColumnDisturb while incurring zero performance and power overhead.

Security contests in the form of CTF (Capture The Flag) exercises are nowadays a common way to learn cyber security. 20 years ago at DIMVA 2006 the on-site CTF CIPHER II was one of the conference highlights and led to the foundation of the team ENOFLAG. In this poster, we reflect on the changes in the CTF gameplay and report on lessons learned while running an academic CTF team for 20 years.