Omer Hofman

Oren Rachmil, Roy Betser, Itay Gershon et al.

Aligning proprietary large language models (LLMs) with internal organizational policies has become an urgent priority as organizations increasingly deploy LLMs in sensitive domains such as legal support, finance, and medical services. Beyond generic safety filters, enterprises require reliable mechanisms to detect policy violations within their regulatory and operational frameworks, where breaches can trigger legal and reputational risks. Existing content moderation frameworks, such as guardrails, remain largely confined to the safety domain and lack the robustness to capture nuanced organizational policies. LLM-as-a-judge and fine-tuning approaches, though flexible, introduce significant latency and lack interpretability. To address these limitations, we propose a training-free and efficient method that treats policy violation detection as an out-of-distribution (OOD) detection problem. Inspired by whitening techniques, we apply a linear transformation to decorrelate the model's hidden activations and standardize them to zero mean and unit variance, yielding near-identity covariance matrix. In this transformed space, we use the Euclidean norm as a compliance score to detect policy violations. The method requires only the policy text and a small number of illustrative samples, which makes it light-weight and easily deployable. On a challenging policy benchmark, our approach achieves state-of-the-art results, surpassing both existing guardrails and fine-tuned reasoning models. This work provides organizations with a practical and statistically grounded framework for policy-aware oversight of LLMs, advancing the broader goal of deployable AI governance. Code is available at: https://tinyurl.com/policy-violation-detection

Omer Hofman, Amit Giloni, Yarin Hayun et al.

Object detection models, which are widely used in various domains (such as retail), have been shown to be vulnerable to adversarial attacks. Existing methods for detecting adversarial attacks on object detectors have had difficulty detecting new real-life attacks. We present X-Detect, a novel adversarial patch detector that can: i) detect adversarial samples in real time, allowing the defender to take preventive action; ii) provide explanations for the alerts raised to support the defender's decision-making process, and iii) handle unfamiliar threats in the form of new attacks. Given a new scene, X-Detect uses an ensemble of explainable-by-design detectors that utilize object extraction, scene manipulation, and feature transformation techniques to determine whether an alert needs to be raised. X-Detect was evaluated in both the physical and digital space using five different attack scenarios (including adaptive attacks) and the COCO dataset and our new Superstore dataset. The physical evaluation was performed using a smart shopping cart setup in real-world settings and included 17 adversarial patch attacks recorded in 1,700 adversarial videos. The results showed that X-Detect outperforms the state-of-the-art methods in distinguishing between benign and adversarial scenes for all attack scenarios while maintaining a 0% FPR (no false alarms) and providing actionable explanations for the alerts raised. A demo is available.

Lidor Erez, Omer Hofman, Tamir Nizri et al.

Automated LLM vulnerability scanners are increasingly used to assess security risks by measuring different attack type success rates (ASR). Yet the validity of these measurements hinges on an often-overlooked component: the evaluator who determines whether an attack has succeeded. In this study, we demonstrate that commonly used open-source scanners exhibit measurement instability that depends on the evaluator component. Consequently, changing the evaluator while keeping the attacks and model outputs constant can significantly alter the reported ASR. To tackle this problem, we present a two-phase, reliability-aware evaluation framework. In the first phase, we quantify evaluator disagreement to identify attack categories where ASR reliability cannot be assumed. In the second phase, we propose a verification-based evaluation method where evaluators are validated by an independent verifier, enabling reliability assessment without relying on extensive human annotation. Applied to the widely used Garak scanner, we observe that 22 of 25 attack categories exhibit evaluator instability, reflected in high disagreement among evaluators. Our approach raises evaluator accuracy from 72% to 89% while enabling selective deployment to control cost and computational overhead. We further quantify evaluator uncertainty in ASR estimates, showing that reported vulnerability scores can vary by up to 33% depending on the evaluator. Our results indicate that the outputs of vulnerability scanners are highly sensitive to the choice of evaluators. Our framework offers a practical approach to quantify unreliable evaluations and enhance the reliability of measurements in automated LLM security assessments.

Jonathan Brokman, Omer Hofman, Oren Rachmil et al.

This report presents a comparative analysis of open-source vulnerability scanners for conversational large language models (LLMs). As LLMs become integral to various applications, they also present potential attack surfaces, exposed to security risks such as information leakage and jailbreak attacks. Our study evaluates prominent scanners - Garak, Giskard, PyRIT, and CyberSecEval - that adapt red-teaming practices to expose these vulnerabilities. We detail the distinctive features and practical use of these scanners, outline unifying principles of their design and perform quantitative evaluations to compare them. These evaluations uncover significant reliability issues in detecting successful attacks, highlighting a fundamental gap for future development. Additionally, we contribute a preliminary labelled dataset, which serves as an initial step to bridge this gap. Based on the above, we provide strategic recommendations to assist organizations choose the most suitable scanner for their red-teaming needs, accounting for customizability, test suite comprehensiveness, and industry-specific use cases.

Omer Hofman, Jonathan Brokman, Oren Rachmil et al.

Agentic AI systems, which build on Large Language Models (LLMs) and interact with tools and memory, have rapidly advanced in capability and scope. Yet, since LLMs have been shown to struggle in multilingual settings, typically resulting in lower performance and reduced safety, agentic systems risk inheriting these limitations. This raises concerns about the accessibility of such systems, as users interacting in languages other than English may encounter unreliable or security-critical agent behavior. Despite growing interest in evaluating agentic AI, existing benchmarks focus exclusively on English, leaving multilingual settings unexplored. To address this gap, we propose MAPS, a multilingual benchmark suite designed to evaluate agentic AI systems across diverse languages and tasks. MAPS builds on four widely used agentic benchmarks - GAIA (real-world tasks), SWE-bench (code generation), MATH (mathematical reasoning), and the Agent Security Benchmark (security). We translate each dataset into eleven diverse languages, resulting in 805 unique tasks and 9,660 total language-specific instances - enabling a systematic analysis of the multilingual effect on AI agents' performance and robustness. Empirically, we observe degradation in both performance and security when transitioning from English to other languages, with severity varying by task and correlating with the amount of translated input. Building on these findings, we provide actionable recommendations to guide agentic AI systems development and assessment under multilingual settings. This work establishes the first standardized evaluation framework for multilingual agentic AI, encouraging future research towards equitable, reliable, and accessible agentic AI. MAPS benchmark suite is publicly available at https://huggingface.co/datasets/Fujitsu-FRE/MAPS

Ariel Fogel, Omer Hofman, Eilon Cohen et al.

Open-weight language models are increasingly used in production settings, raising new security challenges. One prominent threat in this context is backdoor attacks, in which adversaries embed hidden behaviors in language models that activate under specific conditions. Previous work has assumed that adversaries have access to training pipelines or deployment infrastructure. We propose a novel attack surface requiring neither, which utilizes the chat template. Chat templates are executable Jinja2 programs invoked at every inference call, occupying a privileged position between user input and model processing. We show that an adversary who distributes a model with a maliciously modified template can implant an inference-time backdoor without modifying model weights, poisoning training data, or controlling runtime infrastructure. We evaluated this attack vector by constructing template backdoors targeting two objectives: degrading factual accuracy and inducing emission of attacker-controlled URLs, and applied them across eighteen models spanning seven families and four inference engines. Under triggered conditions, factual accuracy drops from 90% to 15% on average while attacker-controlled URLs are emitted with success rates exceeding 80%; benign inputs show no measurable degradation. Backdoors generalize across inference runtimes and evade all automated security scans applied by the largest open-weight distribution platform. These results establish chat templates as a reliable and currently undefended attack surface in the LLM supply chain.

Jonathan Brokman, Amit Giloni, Omer Hofman et al.

Distinguishing between real and AI-generated images, commonly referred to as 'image detection', presents a timely and significant challenge. Despite extensive research in the (semi-)supervised regime, zero-shot and few-shot solutions have only recently emerged as promising alternatives. Their main advantage is in alleviating the ongoing data maintenance, which quickly becomes outdated due to advances in generative technologies. We identify two main gaps: (1) a lack of theoretical grounding for the methods, and (2) significant room for performance improvements in zero-shot and few-shot regimes. Our approach is founded on understanding and quantifying the biases inherent in generated content, where we use these quantities as criteria for characterizing generated images. Specifically, we explore the biases of the implicit probability manifold, captured by a pre-trained diffusion model. Through score-function analysis, we approximate the curvature, gradient, and bias towards points on the probability manifold, establishing criteria for detection in the zero-shot regime. We further extend our contribution to the few-shot setting by employing a mixture-of-experts methodology. Empirical results across 20 generative models demonstrate that our method outperforms current approaches in both zero-shot and few-shot settings. This work advances the theoretical understanding and practical usage of generated content biases through the lens of manifold analysis.

Jonathan Brokman, Amit Giloni, Omer Hofman et al.

Diffusion models, today's leading image generative models, estimate the score function, i.e. the gradient of the log probability of (perturbed) data samples, without direct access to the underlying probability distribution. This work investigates whether the estimated score function can be leveraged to compute higher-order differentials, namely p-Laplace operators. We show here these operators can be employed to identify memorized training data. We propose a numerical p-Laplace approximation based on the learned score functions, showing its effectiveness in identifying key features of the probability landscape. We analyze the structured case of Gaussian mixture models, and demonstrate the results carry-over to image generative models, where memorization identification based on the p-Laplace operator is performed for the first time.

Ben Ganon, Alon Zolfi, Omer Hofman et al.

In recent years, large language models (LLMs) have had great success in tasks such as casual conversation, contributing to significant advancements in domains like virtual assistance. However, they often generate responses that are not aligned with human values (e.g., ethical standards, safety), leading to potentially unsafe or inappropriate outputs. While several techniques have been proposed to address this problem, they come with a cost, requiring computationally expensive training or dramatically increasing the inference time. In this paper, we present DIESEL, a lightweight inference-guidance technique that can be seamlessly integrated into any autoregressive LLM to semantically filter undesired concepts from the response. DIESEL can function either as a standalone safeguard or as an additional layer of defense, enhancing response safety by reranking the LLM's proposed tokens based on their similarity to predefined negative concepts in the latent space. Our evaluation demonstrates DIESEL's effectiveness on state-of-the-art conversational models, even in adversarial jailbreaking scenarios that challenge response safety. We also highlight DIESEL's generalization capabilities, showing that it can be used in use cases other than safety, providing general-purpose response filtering.