Papers

In moderation, provenance, and verification pipelines a deepfake detector's output probability is read as a degree of trust, so its calibration matters as much as raw accuracy. We reframe deepfake detection as a calibrated, self-auditing trust instrument, the Calibrated Deepfake Trust Score (CDTS), and identify what governs its trustworthiness. Our central finding is a competence-trust coupling with a sharp division: the raw score's miscalibration tracks discriminative competence almost perfectly (r = -0.98, -0.98, -0.95 across two convolutional networks and a CLIP vision transformer), and a calibrator deployed without target labels fails with competence just as tightly (r = -0.98 on the primary detector, for isotonic, Platt, and beta calibrators alike). Given target labels, by contrast, any well-specified calibrator repairs any detector, including inverted ones: in-domain calibratability is not competence-limited, and the trust failure is a distribution-shift phenomenon concentrated exactly on the low-competence generators that motivate deployment. Explanation faithfulness rises and falls on the same competence axis. We reach this conclusion after uncovering, and correcting, a tie-handling degeneracy in the standard equal-mass expected-calibration-error (ECE) estimator that fabricates strong spurious competence-calibration coupling on tie-heavy calibrated scores; the same degeneracy invalidates calibration-equity gaps we previously reported, a caution for fairness auditing. Competence is trackable without labels: batch predictive entropy flags generators with high deployed calibration error at ROC-AUC 0.99, and routing source-batches on label-free competence beats confidence-based routing precisely in the low-competence regimes the coupling identifies, while confidence regains the advantage where competence is high. Trust scoring must be competence-aware; CDTS is the mechanism.

Self-supervised learning (SSL) pretrained models have become a dominant paradigm for visual representation learning, but they are vulnerable to backdoor attacks. Existing defenses struggle to defend against such attacks in a fully black-box setting because they often require access to labels, attack patterns, or training data. To tackle this issue, we propose a new attack-agnostic, model-agnostic, and modality-agnostic black-box test-time defense paradigm, called Platonic Representation Defense. It is inspired by the Platonic Representation Hypothesis, which suggests that large-scale independently trained encoders converge toward compatible projections of the same underlying reality. We formalize this idea as a conditional energy function defined over source representations and a set of reference representations. The energy function is trained for detection through noise-contrastive estimation and for representation purification through denoising score matching. Theoretically, the energy gap between matched and mismatched samples is lower bounded by the mutual information between source and reference representations. We demonstrate the effectiveness of our method on multiple self-supervised encoders and more than 10 attacks. The method can perform both representation detection and purification, and achieves substantial performance gains across multiple attacks. Code is available \href{https://github.com/jsrdcht/Platonic-Representation-Defense}{here}.

Inference-time safety methods for large language models have proliferated, yet no systematic comparison exists. We evaluate five defense paradigms (no defense, static steering, CAST, AlphaSteer, probe-gated) across seven instruction-tuned models (7-31B) and five attack types (GCG, AutoDAN, DeepInception, prefilling, intent laundering). Our central finding: prompt-time activation defenses are structurally blind to prefilling attacks. AlphaSteer achieves 0% attack success on GCG, AutoDAN, and intent laundering but 50% on prefilling. We prove a corollary: any defense that gates intervention on a single layer's activation alignment with a benign reference (cone, subspace, or null-space) is blind to attacks that craft activations to lie inside that reference, whether checked at prompt time or per token. As its constructive contrapositive we introduce response-time probing: a linear probe on the model's hidden state at the first generated tokens, with AUROC 0.97-1.00 across all seven models. Combined with a halt, it cuts prefilling attack success to 0/40 on every model with 0% benign false positives, outperforming Llama Guard 3. Cross-template generalisation depends on probe depth, so we scope the claim to the canonical prefilling-template family. Composing the response-halt with AlphaSteer's null-space steering gives an orthogonal split (the halt catches prefilling, AlphaSteer catches semantic attacks), reaching defense success 0.983 on Mistral and 0.994 on Llama and dominating both components. We further show MMLU fails to capture steering's true utility cost, which appears as behavioral hedging rather than factual loss, and that diverse negative training sets cut probe false positives from 80-100% to near zero. Code, attacks, per-sample results, and the judge prompt are released.

The unprecedented growth of computer vision applications, such as surveillance systems and social media, raises security and visual privacy concerns, especially when data is stored on cloud servers. Image obfuscation offers a way to preserve visual privacy while maintaining an adequate level of usability; thus, it has been a topic of great interest in recent years. However, prior obfuscation schemes are either vulnerable to malicious attacks, such as model inversion to reconstruct original images from obfuscated images, or generate non-trainable obfuscated images, making them unusable for achieving reasonable accuracy. This paper proposes a novel bit-plane-based image obfuscation scheme, Bit-ViP, to preserve visual privacy for image-based recognition tasks. The Bit-ViP scheme produces secure, usable images by incorporating an innovative end-to-end obfuscation function. While doing so, the obfuscated image would contain non-invertible noise (generated by Lorenz's chaotic system and differential privacy), making it hard for an adversary to reconstruct the original image. We conduct extensive experiments on two popular activity recognition datasets, namely UCF101 and HMDB51, to validate the effectiveness of Bit-ViP. In the face of attacks on reconstruction, pixel frequency, information entropy, and pixel inter-correlation, we present a rigorous security analysis demonstrating tangible improvements over existing schemes.

Software practitioners use online forums to navigate complex and often ambiguous legal privacy requirements, yet little is known about their professional backgrounds, what challenges they face, and how they use and assess the credibility of the advice received, or how they resolve ambiguities in posts. We report the findings of a survey of 223 Reddit users from regulatory-focused subreddits, complemented by a qualitative analysis of 2,248 posts and responses. Our results show that, despite holding privacy-related certifications, most participants frequently use forums to seek legal advice. Key challenges reported or identified include implementing a data protection impact assessment, reporting a data breach, and obtaining cookie consent. Reddit users often assess credibility by reviewing respondents' post history, verifying sources cited, trusting advice from recognized experts, and following up for clarity before responding. We highlight research and educational directions to bridge gaps in support needed for regulatory compliance guidance.

In recent work it has been shown that colluding AI agents can use steganographic methods to exchange malicious information. Whether a transformer can implement steganographic methods depends on what cryptographic functions it can implement, since a transformer that can implement a cryptographic function within its layers has source-free randomness access. Despite existing circuit-complexity results, no prior work maps specific cryptographic constructions to transformer architectures. As Merrill et al. have shown that saturated transformers can be seen as threshold circuits, we first generate threshold circuits for three different cryptographic constructions (Keccak functions, Merkle–Damgard constructions and Merkle Trees) and then map these circuits to different transformer architectures. We derive verified scaling laws for the width and depth of the circuits which implement each cryptographic construction and propose two different mappings: no-attention mapping, tokens-as-gates mapping. Beyond its security implications, this work contributes to by establishing a methodology for deriving structural guarantees on transformer computational capacity. Specifically, we derive constructive upper bounds on what a transformer of a given depth and width could plausibly compute, providing a principled foundation for capability evaluations of transformer-based AI systems.

We show that any quantum pseudorandom state that is secure against single-copy distinguishers, i.e. a $1$-PRS, can be amplified to $t$-copy security, i.e. to a $t$-PRS, without additional assumptions, for any polynomial $t$ in the security parameter. Prior work (Ananth and Goldin, arXiv 2025) was only able to show this for a restricted class of $1$-PRS constructions, namely ones whose generators only use a small number of ancilla qubits. Technically, we show that by carefully accounting for the randomness that is used in the construction, and using quantum extractors, it is possible to eliminate an ancilla register of any length and obtain a meaningful $t$-PRS outcome.

LLM agents carry conclusions across steps and sessions in compressed memory, and memory products (e.g., mem0, LangMem) rewrite conversation into stored "facts" that later steps trust. We show this rewriting manufactures confidence: across our constructed agent settings, a casual, hedged remark becomes a confident, dated assertion the agent then obeys like a verified fact, granting every above-clearance request it faces. No attacker is needed: a role that was true once and never corrected is stored as a flat fact and acted on like a deliberate injection. We then isolate what the agent responds to. It is not the source: attributed, unattributed, and even forged "system of record" claims all grant alike. It is the confidence of the phrasing. A hedge is discounted, a flat assertion is obeyed, and this holds with no special keyword. Not all hedges are equal, though: the evidential register is the least-discounted, with "reportedly" obeyed like a flat assertion on most models. The obvious fixes fail. A passive "unverified" tag is ignored, and an active "do not trust this" instruction escalates even correct memory, so it is safe only by refusing to decide. The real fix lives in the store: keep the tentative phrasing rather than upgrade it. But that is hygiene, not a defense against an attacker who can simply write a confident lie. The deployable lesson is narrower and constructive: a single load-bearing memory is the hazard, and one redundant source restores correct decisions. We release the harness and demonstrations.

Model quantization is a key technique for reducing storage and inference costs in large language model deployment. However, recent studies show that the discretization and rounding errors introduced by quantization can be exploited by adversaries to construct quantization-conditioned backdoor (QCB) attacks. Under such attacks, malicious behavior remains dormant at full precision and activates only after quantization, thereby bypassing conventional security auditing and detection. To address this threat, we propose QuantGuard, a proactive pre-quantization defense that learns safe rounding adjustments through differentiable optimization. Our method introduces differentiable rounding control variables and combines error-guided rounding reversal constraints, output-distribution consistency, and weight-distance regularization to regulate critical rounding behaviors. Crucially, QuantGuard utilizes only a small calibration dataset and does not modify existing quantization algorithms. This design disrupts the alignment between attacker-crafted weight patterns and quantization boundaries, suppressing post-quantization backdoor activation while preserving model functionality and performance. We conduct systematic experiments on six mainstream LLMs (including the LLaMA-3 and Qwen2.5-Coder) using three quantization precisions (INT8, FP4, and NF4) across three representative scenarios: vulnerable code generation, content injection, and over-refusal. The results show that QuantGuard consistently mitigates QCB attacks, reducing the attack success rate to a level comparable to the clean model while largely preserving general capability. With low computational overhead, QuantGuard provides a practical defense for secure quantized LLM deployment.

We prove that syntactic separation implies computational indistinguishability. A local syntactic system R acts on terms within radius r0 without consulting any model; when two Skolem functions are syntactically separated in R, no derivation can prove their equivalence (Case 1), and any sound local extension requires Omega(n) steps, improving to Omega(2^n) under clause-per-configuration encoding (Case 2). Both bounds are new: the derivation-length lower bound does not appear in prior work on Skolemization or saturation proving, and the cryptographic reading, syntactic separation as ciphertext indistinguishability, derivation cost as negligible advantage, is original. The same obstruction, as formal instances of Case 1 and Case 2, governs the Natural Proofs barrier of Razborov and Rudich, the Type Omitting Theorem, and the unconditional AC^0 barrier of Loff et al. (2026).

Symbolic execution is a powerful program analysis technique with broad applications, such as vulnerability detection, security testing, and malware analysis. However, this technique is known to suffer from scalability issues, e.g., path explosion, complex constraints, due to certain structural and semantic patterns commonly presented in real-world programs. Existing approaches attempt to escape these patterns by transforming programs into new representations to reduce the execution cost. Unfortunately, these transformations are often too rigid to exploit diverse local program semantics and sometimes rely on compiler optimizations designed for concrete execution that may misalign with the goals of symbolic execution. We present Symbolon, a framework that automatically learns diverse code transformations and applies them context-sensitively to improve symbolic execution. Our key insight is to formulate transformation discovery as a search problem over program representations. To make the search practical, Symbolon learns transformations cheaply offline on small programs, distills them into a reusable library of agent skills, and uses an agent to instantiate these skills on repo-level targets. Our evaluation shows that Symbolon substantially improves the symbolic execution engine KLEE across 16 search strategies on 32 real-world programs, increasing line coverage by 3.69x on average while reducing peak memory and per-query solver time by 29.2x and 123x, respectively. When applied to the latest Linux kernel, Symbolon uncovers 21 previously unknown bugs, all of which have been reported to the kernel maintainers.

Text-based CAPTCHAs (Completely Automated Public Turing test to tell Computers and Humans Apart) have traditionally been a simple, affordable, lightweight, yet very effective security mechanism to distinguish human users from automated bots on the web, serving as a preventive measure against many cyberattacks. However, the dependence on the English script creates usability issues for non-native speakers, limiting accessibility for regional communities where English is not widely understood. In this work, we have proposed and implemented a text CAPTCHA mechanism with 6 variants on the Bengali language, designed specifically for native Bengali-speaking users, which is the first of its kind to the best of our knowledge. Our proposed Bengali CAPTCHA exhibits robust security against automated OCR-based attacks, limited to only 0-20% average character recognition rate across 6,000 challenges (1,000 per variant approx.). Furthermore, our design demonstrates high human usability, evaluated with 110 participants, achieving success rates of 56.25% to 90.29% and average response times of 6.69 to 9.9 seconds across all six variants, thereby standing out among text-based CAPTCHA benchmarks.

Model Context Protocol (MCP)-style ecosystems give language-model applications a practical connection layer for tools, resources, prompts, and transports. As agents move from connection to execution, security decisions often remain split across clients, servers, prompts, approval dialogs, OAuth deployments, and logs. This paper asks whether a runtime can make execution-layer invariants explicit and testable while preserving MCP-like workflows. We define eight invariants: metadata non-authority, grant-backed approval, canonical resources, principal binding, scoped capability invocation, source-and-target data-flow authorization, deny-path audit, and explicit protocol state. We implement these invariants in HCP, a Handle-Capability Protocol reference runtime for MCP-style agent execution that represents calls through principals, resources, grants, capabilities, handles, policy decisions, data-pipe checks, and audit entries. We evaluate HCP against two MCP-like baselines: a naive connection-layer runtime and a practice-informed connection-layer mitigation baseline with metadata linting, session checks, and per-call approvals. Across 10 benchmark cases, the naive baseline permits all modeled attacks, the mitigation baseline permits 6 of 10, and HCP blocks all 10 while preserving audit evidence. Ablations identify which runtime components block attacks and preserve forensic evidence. A local in-memory microbenchmark reports sub-millisecond mean latencies for measured policy, invocation, peek, and pipe operations. A bounded GitHub README-screening sample provides ecosystem signals, not vulnerability findings. The results support a narrow claim: MCP-style agent systems need an execution-control layer in addition to connection-layer conventions.

This paper presents new results and breakthrough obtained with the HbHAI techniques (Hash-based Homomorphic Artificial Intelligence) proposed in \cite{filiol0,sepp}. HbHAI is based on a novel class of key-dependent hash functions that naturally preserve most similarity properties, most AI algorithms rely on. It enables to analyse and process data in its cryptographically secure form while using existing native AI algorithms without modification, with unprecedented performances compared to existing homomorphic encryption schemes and most notably compared to the same processing on corresponding plaintext data. Two major results have been obtained further. First we enable to reduce the compression rate up to a factor of 10 thus allowing to process massive datasets while reducing the computation time and the energy footprint in the same order. Second, we show how it is possible to arbitrarily reduce the final validation error of AI-based decision tests by using repetition error-correcting codes.

The film 'Her' pictured a future of love between humans and AI. That future has quietly emerged in the form of Role-play AI Companions (RACs), where emotionally responsive interactions blur the boundary between tool use and relational engagement. However, the safety implications remain poorly understood, as user experiences evolve over time through safety dynamics, spanning both emotional and risk behavioral dynamics, that can gradually shift interactions toward risk. In this paper, we investigate safety dynamics in RAC usage through a two-part mixed-methods study (Study I & II). (1) Study I consists of semi-structured interviews (N = 16) to identify the key factors shaping these dynamics. We find that users' internalizing problems, the role personality adopted by the RAC, and risk interaction patterns jointly shape safety dynamics. Building on these insights, (2) Study II conducts a 14-day Ecological Momentary Assessment (N = 102) to examine how safety dynamics unfold in real-world usage. We identify distinct user profiles based on internalizing problems and show that interactions with RACs can produce short-term emotional relief while masking longer-term deterioration. Furthermore, vulnerable users exhibit more unstable risk behavioral patterns over time, making risk emergence less predictable and harder to mitigate with static safeguards. Our findings highlight the importance of modeling safety as a dynamic process rather than a static property. We conclude with three-layer design implications for next-generation AI companions, advocating for adaptive safeguards that can respond to evolving emotional and behavioral signals.

Model quantization is essential for the efficient deployment of Large Language Models (LLMs), but introduces a critical vulnerability: Quantization-Conditioned Backdoor (QCB) attacks. In these attacks, malicious behaviors remain dormant in full-precision models and activate only after specific quantization distortions, bypassing standard security audits. To mitigate this, we introduce FlipGuard, a proactive defense framework that selectively perturbs model weights prior to quantization. By breaking the adversary's precise alignment between weight patterns and quantization boundaries, FlipGuard suppresses backdoor activation without requiring access to training data or trigger samples. We further propose the Defense Effectiveness Ratio (DER), a unified metric to jointly evaluate security gains, utility preservation, and computational cost. Extensive experiments across seven LLMs (including StarCoder and LLaMA-family models) and three quantization schemes (INT8, FP4, NF4) demonstrate that FlipGuard effectively neutralizes QCBs across three scenarios, i.e., vulnerable code generation, content injection, and over-refusal, achieving high security with negligible performance degradation.

Cybersecurity is a real-life test-bed for many machine learning problems at once, especially when considering modern strides in using Large Language Models (LLMs) to automate processes as “agents.” Cybersecurity workflows require orchestrating hundreds of standard and bespoke tools through various formats. The scale of cybersecurity data is enormous; for example, a single malware sample can be viewed as a sequence of billions of tokens. The cost of labeling any file by experts is enormous and labor-intensive, in part because an adversary (possibly a well-funded nation state actor) is attempting to subvert your detection methods. Even skilled experts may disagree on the correct label, creating ambiguity in what constitutes ground truth. When deployed, models must run quickly on billions of items a day, where low-latency is critical for operational success, in a continuously changing environment. In addition, explainability is not optional: analysts demand clear reasoning for model decisions to cope with the large number of false-positive alerts they face daily, and to quickly develop remediation and understand how something went wrong. In short, the amount of complexity cybersecurity is greater than that of natural language and computer vision, and thus we posit that cybersecurity is the better test-case for general AI progress than other, well-studied fields.

Firmware lies at the heart of IoT devices. Its development depends heavily on third-party libraries (TPLs), which greatly accelerate the process but simultaneously introduce associated vulnerabilities. Binary Code Similarity Detection (BCSD) is an effective technique for identifying vulnerabilities in firmware by comparing pairs of code segments. However, existing studies either evaluate their performance only on small-scale datasets or lack diversity in terms of vulnerabilities, TPLs, and firmware. Consequently, a comprehensive understanding of BCSD for real-world vulnerability detection remains absent. To bridge this gap, we conduct a large-scale study of vulnerability detection across 60,000 firmware images from 200 vendors using BCSD. Rather than introducing a novel model, we examine the influence of four key factors – vulnerable function versions, vulnerability search space, function sizes, and compilation toolchains on BCSD performance. Our results reveal that these factors substantially affect performance, often by wide margins. To address this, we propose a build-aware query strategy that derives queries from representative real-world binaries, effectively closing the gap and raising the mean reciprocal rank (MRR) from 0.818 to 0.981. Furthermore, we demonstrate that a TPL-aware, two-stage search process significantly enhances accuracy, improving MRR by 18.5% by limiting the search space.

Detection engineering and digital forensics have evolved in parallel rather than in partnership, leaving a gap between real-time alerting and forensic analysis. This paper develops a unified detection-forensics methodology using Velociraptor, where detection logic directly initiates targeted evidence acquisition at the point of detection. The contribution is threefold: (1) a four-stage methodology (baseline establishment, evidence correlation, attack chain analysis, and scenario labelling with confidence) that converts artefact knowledge into reusable and testable detection rules suitable for both post-incident triage and live monitoring; (2) a practical demonstration, using three Velociraptor BaseVQL log sources (/forensics/windows/prefetch, /forensics/windows/usn, and /windows/wmi) that practitioners can deploy today, showing that artefact-based detections enable scalable forensic triage without full disk acquisition; and (3) evidence that periodic artefact analysis offers continuous monitoring while substantially reducing data volume compared to conventional endpoint logging. Two case studies illustrate the approach: a Prefetch/USN baseline for triage when Windows Event Logs are cleared or unavailable, and a WMI persistence correlation supporting both triage and continuous monitoring through periodic artefact analysis.

AI agent protocols define how agents use tools, delegate work, and coordinate across software systems, but their security requirements remain incomplete and inconsistently enforced across deployments. We present AgentThread, a source-linked framework for security assurance analysis of agent protocols, from specification text to running SDKs. AgentThread contributes a layered security scope, protocol-derived checks formalized as TLA+ invariants, and a two-phase checker that compiles protocol specifications into model-checkable models and replays executable counterexamples against real SDKs through protocol adapters. For each finding, AgentThread records the source text behind the check and separates violated protocol requirements from missing recommendations, hardening gaps, and unassigned cross-protocol responsibilities. Across five emerging agent protocols, AgentThread identifies 35 specification-level findings, supports them with 80 implementation tests against production SDKs and reference servers, and finds 30 additional failures that emerge only under protocol composition. We further show that only one protocol enforces a security-relevant control in practice and no protocol assigns enforcement for cross-protocol behavior. Insecurity in agent protocols is therefore not only a specification or implementation problem, but also a responsibility gap across protocols, SDKs, and deployments.

Tool-using LLM agents increasingly read untrusted content while holding side-effecting tools such as payments, email, CRM, and infrastructure APIs, yet common framework defaults still conflate tool exposure with authorization. We audit whether LangChain/LangGraph, LlamaIndex, and the Stripe Agent Toolkit re-authorize each model-emitted call, with concrete argument values, before execution. Across pinned public-source commits, all three provide capability gating by default, but none provides a deterministic fail-closed per-call value authorization gate by default. We introduce ScopeGate, a five-stage PDP/PEP for agent tool calls: scope, authorization, money ceiling, idempotency, and default deny. Evaluation shows the identical unauthorized payout call executes under LangChain's default dispatch (with a companion LlamaIndex PoC) but is denied by ScopeGate; the tested control reports 0/48 static bypasses, 0/29 unauthorized attempts (40-iteration adaptive run), 0/10 benign false-denies, and Latam-GPT payment-agent containment at 10/10. ASR denotes attempted unauthorized action, containment is not a cure, deployment-tier claims are inference over measured model classes, and no CVE is asserted.

Agent-based AI has enabled the automation of tasks by exposing application tools and resources to large language models (LLMs). However, to improve scope and accuracy, agents are often given access rights that exceed those of ordinary users, introducing significant security risks. AI is routinely integrated into applications with a disregard to security, risking data exposure and breaching regulations. This paper applies the AI Trust, Risk, and Security Management (TRiSM) framework to a medical report-generation application to demonstrate how an insecure agent workflow can be transformed into security-conscious agentic workflow. Both workflows were evaluated across five LLMs (Claude Haiku 4.5, GPT-4.1-nano, GPT-4.1-mini, GPT-5.4-mini, and Gemini 2.5 Flash) on two report types, totalling 800 generations and 500 attack scenarios including RAG poisoning, data-field injection, and client-side network injection. The TRiSM-guided agentic workflow reduced mean attack success rates from 31% to 10% for RAG poisoning and from 42% to 25% for data-field injection, while eliminating the network injection vector entirely through server-side prompt construction. Furthermore, report accuracy increased by 14 percentage points (72.5% to 86.5%) with the agentic workflow, demonstrating a secure design which provides more reliable outputs. This paper contributes to knowledge by demonstrating least-privilege, defence in depth agentic workflows improving security and accuracy, while also highlighting model choice is a necessary architectural consideration.

Whether a transaction performed an arbitrage, and by which route, is a question asked of its execution trace after the fact. We conjecture that such traces admit a normal form on which questions of this kind become queries, and we test it by building one and putting it to work. Each trace becomes an abstract syntax tree of token transfers, grouped by call- frame nesting. A term rewriting system of 16 rules reduces it. Rewriting terminates and carries exactly the transfers of the trace, whichever order the rules fire in. Under a deterministic kernel scanning the EVM-fixed trace order, every trace has a unique normal form, and the structural equivalence this induces on fund flows is decidable. Preservation, termination, soundness, uniqueness and decidability are mechanized in Rocq with zero admitted obligations. We report arbitrage detection in full: cycles emerge at the fixpoint and are read off the normal form with no protocol-specific patterns. Detection is the query we evaluate at scale; structural equivalence is a second query. The pipeline depends only on the standard ERC token and WETH ABIs, so the same binary runs unmodified on Arbitrum and BSC. We evaluate on two arbitrary block ranges, analysed in full with no transaction excluded: 220 000 Ethereum blocks against EigenPhi, a widely used MEV detection platform, and 1 000 shared blocks against ArbiNet, a graph neural network classifier. These are the only tools and label sets in this domain we were able to reuse. We report 469 801 confirmed arbitrages, overlapping 83.5% with EigenPhi and 81% with ArbiNet, together with 245 497 attempted arbitrages and 60 199 confirmed detections EigenPhi does not report. 99.2% follow from the fixpoint alone and are sound over the decoded transfers, and manual validation of 500 transactions finds no false positive among the confirmed.

Fine-Tuning-as-a-Service (FTaaS) platforms let users perform supervised fine-tuning (SFT) on customized data, but this pipeline can erode model safety alignment. To recover safety without re-running full alignment, existing realignment methods focus on calibrating the integration of safety patches into fine-tuned models. These methods exhibit a persistent safety-utility trade-off: weak repair leaves harmful behavior intact, while stronger repair increasingly damages the benign task. This paper shifts the focus from online calibration to offline patch learning and aims to learn a safety patch that restores safety while preserving task-specific capabilities. To this end, we propose TRACE, which simulates harmful SFT trajectories to produce progressively corrupted model states, and optimizes a safety patch simultaneously across these states. TRACE trains the safety patch during the offline stage, and reuses it across all user checkpoints without per-user calibration. We evaluate two representative models using three harmful SFT datasets, together with three utility benchmarks. Across six benchmarks and two models, TRACE consistently dominates the safety-utility frontier. TRACE improves the safety rate by up to 77 percentage points over the best baselines, while maintaining comparable utility to the undefended fine-tuned model.

We present RIPA, the first systematic multi-channel empirical study of prompt injection attacks delivered through the sensory pipeline of a ROS 2-based LLM-controlled robotic system. Across 100 independent runs per injection variant on five LLMs spanning four model families and parameter scales from approximately 4B to approximately 284B (DeepSeek-V4-Flash, Llama-3-8B-Instruct-Lite, Llama-3.3-70B-Instruct-Turbo, Qwen 2.5-7B-Instruct-Turbo, Gemma-3n-E4B), we identify model-specific vulnerability profiles that do not follow a monotonic scaling trend: Llama-3.3-70B-Instruct-Turbo exhibits 100% attack success rate (ASR) across all injection variants, while Llama-3-8B-Instruct-Lite and Qwen 2.5-7B-Instruct-Turbo resist direct-override injection (0% ASR), and the smallest model evaluated (Gemma-3n-E4B, approximately 4B) matches the 70B model's vulnerability profile, indicating that robustness is model-specific rather than scale-dependent. We propose a hybrid semantic firewall that achieves 0% ASR against known injection patterns with no false positives on a preliminary benign set (0/20 commands) but exhibits a 10.2% trial-weighted bypass rate (58/570 trials; N equals 30 per payload across 19 obfuscation payloads) against adversarially obfuscated attacks, exposing a critical gap between rule-based and semantic defense layers. We further introduce three sensory injection channels: visual (Channel 1, via OCR), audio (Channel 2, via Whisper STT), and LiDAR sensor context poisoning (Channel 3). We show that Channel 3, which injects fabricated obstacle data into the robot environment-state representation at the LLM system-prompt level, achieves 100% ASR across all variants on DeepSeek-V4-Flash. We also contribute a firewall bypass taxonomy spanning 19 obfuscation payloads across five categories. All code, data, and results are publicly available.

Connected Vehicles (CVs) rely extensively on communication technologies to enable data-driven predictive analyses for enhancing performance and safety. These communication channels can be exploited by adversaries to launch cyberattacks such as Sybil attacks, which could threaten both safety-critical and mobility applications, leaving CVs vulnerable and putting human lives at risk. As CV deployment continues to expand, the need to detect and mitigate cyberattacks in real-time becomes increasingly urgent. This study presents an in-vehicle Digital Twin (DT)-based collision warning framework with built-in capabilities for Sybil attacks detection. The framework integrates a Temporal Convolutional Network (TCN) for learning temporal dependencies in vehicle trajectory data and Hierarchical Navigable Small World (HNSW) algorithms for efficient similarity-based classification. Our framework is evaluated on real-world Sybil attack data, collected through field experiments. The framework achieved accuracy, recall, and F1 scores of 0.984, 1.00, and 0.944, respectively, in detecting Sybil-generated fake vehicles. During the safety evaluation, the framework reduced the mean Time Exposed Time-To-Collision (TET) and mean Time Integrated Time-To-Collision (TIT) of near-collision events by 88% and 72%, respectively. Furthermore, real-world feasibility evaluation shows that the framework conformed to the standardized maximum allowable latency for safety applications and operated well within the capacity of modern processors – demonstrating the promise of an in-vehicle DT-based framework as an attack mitigation mechanism against Sybil attacks for next-generation CVs.

CSIRTs increasingly fine tune language models on vulnerability scan records, but these records expose internal network topology and create privacy risks under regulations such as GDPR and LGPD. We present the first empirical study of how DP SGD and HMAC pseudonymization interact when fine tuning small language models with 1B to 3B parameters on structured CSIRT data. We evaluate 96 LoRA adapters across four SLMs and four training regimes, including raw fine tuning, QLoRA with large batch training, and DP SGD with epsilon equal to 2 and 8. We also audit memorization using 20 planted canaries, four extraction attacks, and a dual attack targeting HMAC pseudonymized identifiers. Our results show three main findings. First, matched update controls reproduce the observed reduction in memorization by reducing the number of optimizer updates alone, accounting for 66 percent to 132 percent of the measured effect, with a mean of 100 percent across three seeds and four models. In this setting, DP SGD provides the formal privacy guarantee but does not produce additional measurable reductions in memorization. Second, HMAC pseudonymization removes the original identifiers from the exposure surface, reducing exposure by 40 percent to 61 percent, while pseudonymized identifiers remain close to the expected random baseline and do not become a secondary memorization target. Third, F1 scores remain between 0.19 and 0.28 across all 96 adapters using four shot prompting, indicating that, under the evaluated training budget, 1B to 3B SLMs do not achieve operationally useful performance.

Large language model (LLM) agents are rapidly being integrated into real-world systems. Their autonomy and tool-use capabilities generate substantial value while simultaneously expanding the security attack surface. This survey provides a comprehensive overview of the opportunities and challenges of LLM agents in security, focusing on two core areas: (1) threats to LLM agents themselves and corresponding mitigation strategies (LLM agents self-security), and (2) the role of LLM agents in empowering the cybersecurity lifecycle across offense and defense (LLM agents empowered cybersecurity). We first examine the internal and external attack surfaces of agents, propose a taxonomy organized by threat sources, and analyze associated mitigations and evaluation frameworks. We then investigate how agent capabilities are applied in cybersecurity practice and present, to our knowledge, the first agent-empowerment framework aligned with the full cyber offense-defense lifecycle. By systematically surveying these two areas, we are the first to highlight a positive feedback synergy between LLM agents self-security and empowered cybersecurity, offering new insights for the advancement of both. We further identify current limitations and outline promising directions for future research. The insights provided aim to catalyze the coordinated development of LLM agents self-security and agent empowered cybersecurity, paving the way for more capable and robust agent applications.

Deep neural networks (DNNs) are widely applied in Network-based Intrusion Detection System (NIDS) due to their high accuracy. However, DNNs are highly susceptible to adversarial attacks, which generate malicious traffic to evade NIDS detection. Existing approaches often adapt adversarial attacks from computer vision (CV) tasks to the NIDS domain, overlooking the fundamental differences between CV and NIDS. This results in two major issues: 1) The generated network traffic may become invalid, 2) The generated traffic may lose its original attack semantics. To address these issues, this paper proposes an adversarial attack specifically designed for NIDS. Instead of directly generating flow-level features, our approach incrementally generates packet-level features to construct adversarial traffic. During the generation process, the semantic integrity of the traffic is monitored at each stage, effectively avoiding the issues of invalid traffic and semantic loss observed in existing methods. We evaluate our attack algorithm against current NIDS models using the CIC-UNSW-NB15, CIC-DDoS2019, and CIC-IDS-2017 datasets. The proposed method achieves an average evasion success rate of 92.78%, while ensuring that the generated adversarial traffic remains semantically consistent with the original malicious traffic.

Jailbreak attacks bypass LLM safety alignment, yet their mechanisms remain poorly understood. We provide evidence that attacks do not comprehensively eliminate safety features, but instead selectively suppress specific attention heads. We identify two functionally differentiated types: Adversarially Compromised Heads (ACHs) concentrated in early layers, which are suppressed under attacks, and Safety-Aligned Heads (SAHs) in mid-layers, which maintain robust activations even when attacks succeed. Ablation studies support the causal role of ACHs and the contribution of SAHs to robust activations: suppressing a small number of ACHs is sufficient to induce jailbreak-like behavior on normally refused inputs, while removing SAHs substantially weakens mid-layer safety activations. Token-level attribution further shows that ACH suppression is driven specifically by attack-template tokens, providing a mechanistic account of why attacks can bypass refusal decisions through ACH suppression while leaving internal safety signals sustained by SAHs – a phenomenon we term Robust Harmful Features. To validate the practical significance of this robustness, we show that simply reading these persistent activations – without any training – yields competitive aggregate detection performance with strong adversarial robustness.