3D Gaussian Splatting (3DGS) is increasingly recognized as a powerful paradigm for real-time, high-fidelity 3D reconstruction. However, its per-scene optimization pipeline limits scalability and generalization, and prevents efficient inference. Recently emerged feed-forward 3DGS models address these limitations by enabling fast reconstruction from a few input views after large-scale pretraining, without scene-specific optimization. Despite their advantages and strong potential for commercial deployment, the use of neural networks as the backbone also amplifies the risk of adversarial manipulation. In this paper, we introduce AdvSplat, the first systematic study of adversarial attacks on feed-forward 3DGS. We first employ white-box attacks to reveal fundamental vulnerabilities of this model family. We then develop two improved, practically relevant, query-efficient black-box algorithms that optimize pixel-space perturbations via a frequency-domain parameterization: one based on gradient estimation and the other gradient-free, without requiring any access to model internals. Extensive experiments across multiple datasets demonstrate that AdvSplat can significantly disrupt reconstruction results by injecting imperceptible perturbations into the input images. Our findings surface an overlooked yet urgent problem in this domain, and we hope to draw the community's attention to this emerging security and robustness challenge.

Genomic language models (GLMs) have emerged as powerful tools for learning representations of DNA sequences, enabling advances in variant prediction, regulatory element identification, and cross-task transfer learning. However, as these models are increasingly trained or fine-tuned on sensitive genomic cohorts, they risk memorizing specific sequences from their training data, raising serious concerns around privacy, data leakage, and regulatory compliance. Despite growing awareness of memorization risks in general-purpose language models, little systematic evaluation exists for these risks in the genomic domain, where data exhibit unique properties such as a fixed nucleotide alphabet, strong biological structure, and individual identifiability. We present a comprehensive, multi-vector privacy evaluation framework designed to quantify memorization risks in GLMs. Our approach integrates three complementary risk assessment methodologies: perplexity-based detection, canary sequence extraction, and membership inference. These are combined into a unified evaluation pipeline that produces a worst-case memorization risk score. To enable controlled evaluation, we plant canary sequences at varying repetition rates into both synthetic and real genomic datasets, allowing precise quantification of how repetition and training dynamics influence memorization. We evaluate our framework across multiple GLM architectures, examining the relationship between sequence repetition, model capacity, and memorization risk. Our results establish that GLMs exhibit measurable memorization and that the degree of memorization varies across architectures and training regimes. These findings reveal that no single attack vector captures the full scope of memorization risk, underscoring the need for multi-vector privacy auditing as a standard practice for genomic AI systems.

Large language models (LLMs) often require fine-tuning (FT) to perform well on downstream tasks, but FT can induce safety-alignment drift even when the training dataset contains only benign data. Prior work shows that introducing a small fraction of harmful data can substantially compromise LLM refusal behavior, causing LLMs to comply with harmful requests. Existing defense methods often rely on model-wide interventions, such as restricting which parameters are updated or injecting additional safety data, which can limit generality and degrade downstream task performance. To address these limitations, we propose a fine-tuning framework called Preserving Safety Alignment via Constrained Tokens (PACT), which stabilizes the model's confidence on safety tokens. Our approach is motivated by the empirical observation that safety-aligned behavior is reflected in the model's token-level output confidence and is often concentrated on a small subset of safety-related tokens. During downstream fine-tuning, we regularize the fine-tuned model to match the aligned reference model's confidence on safety-related tokens at each response step, while leaving non-safety tokens largely unconstrained to allow effective task adaptation. This targeted constraint prevents alignment drift without imposing global restrictions that typically trade off with model utility.

Cryptographic provenance standards such as C2PA and invisible watermarking are positioned as complementary defenses for content authentication, yet the two verification layers are technically independent: neither conditions on the output of the other. This work formalizes and empirically demonstrates the $\textit{Integrity Clash}$, a condition in which a digital asset carries a cryptographically valid C2PA manifest asserting human authorship while its pixels simultaneously carry a watermark identifying it as AI-generated, with both signals passing their respective verification checks in isolation. We construct metadata washing workflows that produce these authenticated fakes through standard editing pipelines, requiring no cryptographic compromise, only the semantic omission of a single assertion field permitted by the current C2PA specification. To close this gap, we propose a cross-layer audit protocol that jointly evaluates provenance metadata and watermark detection status, achieving 100% classification accuracy across 3,500 test images spanning four conflict-matrix states and three realistic perturbation conditions. Our results demonstrate that the gap between these verification layers is unnecessary and technically straightforward to close.

3D Gaussian Splatting (3DGS) has emerged as a powerful paradigm for real-time and high-fidelity 3D reconstruction from posed images. However, recent studies reveal its vulnerability to adversarial corruptions in input views, where imperceptible yet consistent perturbations can drastically degrade rendering quality, increase training and rendering time, and inflate memory usage, even leading to server denial-of-service. In our work, to mitigate this issue, we begin by analyzing the distinct behaviors of adversarial perturbations in the low- and high-frequency components of input images using wavelet transforms. Based on this observation, we design a simple yet effective frequency-aware defense strategy that reconstructs training views by filtering high-frequency noise while preserving low-frequency content. This approach effectively suppresses adversarial artifacts while maintaining the authenticity of the original scene. Notably, it does not significantly impair training on clean data, achieving a desirable trade-off between robustness and performance on clean inputs. Through extensive experiments under a wide range of attack intensities on multiple benchmarks, we demonstrate that our method substantially enhances the robustness of 3DGS without access to clean ground-truth supervision. By highlighting and addressing the overlooked vulnerabilities of 3D Gaussian Splatting, our work paves the way for more robust and secure 3D reconstructions.

Membership inference attacks (MIAs) threaten the privacy of machine learning models by revealing whether a specific data point was used during training. Existing MIAs often rely on impractical assumptions such as access to public datasets, shadow models, confidence scores, or training data distribution knowledge and making them vulnerable to defenses like confidence masking and adversarial regularization. Label-only MIAs, even under strict constraints suffer from high query requirements per sample. We propose a cost-effective label-only MIA framework based on transferability and model extraction. By querying the target model M using active sampling, perturbation-based selection, and synthetic data, we extract a functionally similar surrogate S on which membership inference is performed. This shifts query overhead to a one-time extraction phase, eliminating repeated queries to M . Operating under strict black-box constraints, our method matches the performance of state-of-the-art label-only MIAs while significantly reducing query costs. On benchmarks including Purchase, Location, and Texas Hospital, we show that a query budget equivalent to testing $\approx1\%$ of training samples suffices to extract S and achieve membership inference accuracy within $\pm1\%$ of M . We also evaluate the effectiveness of standard defenses proposed for label-only MIAs against our attack.

In recent years, the widespread adoption of Machine Learning as a Service (MLaaS), particularly in sensitive environments, has raised considerable privacy concerns. Of particular importance are membership inference attacks (MIAs), which exploit behavioral discrepancies between training and non-training data to determine whether a specific record was included in the model's training set, thereby presenting significant privacy risks. Although existing defenses, such as adversarial regularization, DP-SGD, and MemGuard, assist in mitigating these threats, they often entail trade-offs such as compromising utility, increased computational requirements, or inconsistent protection against diverse attack vectors. In this paper, we introduce a novel inference-time defense mechanism called Neighborhood Blending, which mitigates MIAs without retraining the model or incurring significant computational overhead. Our approach operates post-training by smoothing the model's confidence outputs based on the neighborhood of a queried sample. By averaging predictions from similar training samples selected using differentially private sampling, our method establishes a consistent confidence pattern, rendering members and non-members indistinguishable to an adversary while maintaining high utility. Significantly, Neighborhood Blending maintains label integrity (zero label loss) and ensures high utility through an adaptive, "pay-as-you-go" distortion strategy. It is a model-agnostic approach that offers a practical, lightweight solution that enhances privacy without sacrificing model utility. Through extensive experiments across diverse datasets and models, we demonstrate that our defense significantly reduces MIA success rates while preserving model performance, outperforming existing post-hoc defenses like MemGuard and training-time techniques like DP-SGD in terms of utility retention.

Current approaches to LLM safety fundamentally rely on a brittle cat-and-mouse game of identifying and blocking known threats via guardrails. We argue for a fresh approach: robust safety comes not from enumerating what is harmful, but from deeply understanding what is safe. We introduce Trust The Typical (T3), a framework that operationalizes this principle by treating safety as an out-of-distribution (OOD) detection problem. T3 learns the distribution of acceptable prompts in a semantic space and flags any significant deviation as a potential threat. Unlike prior methods, it requires no training on harmful examples, yet achieves state-of-the-art performance across 18 benchmarks spanning toxicity, hate speech, jailbreaking, multilingual harms, and over-refusal, reducing false positive rates by up to 40x relative to specialized safety models. A single model trained only on safe English text transfers effectively to diverse domains and over 14 languages without retraining. Finally, we demonstrate production readiness by integrating a GPU-optimized version into vLLM, enabling continuous guardrailing during token generation with less than 6% overhead even under dense evaluation intervals on large-scale workloads.

Split Learning (SL) offers a framework for collaborative model training that respects data privacy by allowing participants to share the same dataset while maintaining distinct feature sets. However, SL is susceptible to backdoor attacks, in which malicious clients subtly alter their embeddings to insert hidden triggers that compromise the final trained model. To address this vulnerability, we introduce SecureSplit, a defense mechanism tailored to SL. SecureSplit applies a dimensionality transformation strategy to accentuate subtle differences between benign and poisoned embeddings, facilitating their separation. With this enhanced distinction, we develop an adaptive filtering approach that uses a majority-based voting scheme to remove contaminated embeddings while preserving clean ones. Rigorous experiments across four datasets (CIFAR-10, MNIST, CINIC-10, and ImageNette), five backdoor attack scenarios, and seven alternative defenses confirm the effectiveness of SecureSplit under various challenging conditions.

Small language models (SLMs) are increasingly deployed on edge devices, making their safety alignment crucial yet challenging. Current shallow alignment methods that rely on direct refusal of malicious queries fail to provide robust protection, particularly against adversarial jailbreaks. While deliberative safety reasoning alignment offers deeper alignment for defending against sophisticated attacks, effectively implanting such reasoning capability in SLMs with limited capabilities remains an open challenge. Moreover, safety reasoning incurs significant computational overhead as models apply reasoning to nearly all queries, making it impractical for resource-constrained edge deployment scenarios that demand rapid responses. We propose EASE, a novel framework that enables practical and Efficient safety Alignment for Small languagE models. Our approach first identifies the optimal safety reasoning teacher that can effectively distill safety reasoning capabilities to SLMs. We then align models to selectively activate safety reasoning for dangerous adversarial jailbreak queries while providing direct responses to straightforward malicious queries and general helpful tasks. This selective mechanism enables small models to maintain robust safety guarantees against sophisticated attacks while preserving computational efficiency for benign interactions. Experimental results demonstrate that EASE reduces jailbreak attack success rates by up to 17% compared to shallow alignment methods while reducing inference overhead by up to 90% compared to deliberative safety reasoning alignment, making it practical for SLMs real-world edge deployments.

This study investigates embedding reconstruction attacks in large language models (LLMs) applied to genomic sequences, with a specific focus on how fine-tuning affects vulnerability to these attacks. Building upon Pan et al.'s seminal work demonstrating that embeddings from pretrained language models can leak sensitive information, we conduct a comprehensive analysis using the HS3D genomic dataset to determine whether task-specific optimization strengthens or weakens privacy protections. Our research extends Pan et al.'s work in three significant dimensions. First, we apply their reconstruction attack pipeline to pretrained and fine-tuned model embeddings, addressing a critical gap in their methodology that did not specify embedding types. Second, we implement specialized tokenization mechanisms tailored specifically for DNA sequences, enhancing the model's ability to process genomic data, as these models are pretrained on natural language and not DNA. Third, we perform a detailed comparative analysis examining position-specific, nucleotide-type, and privacy changes between pretrained and fine-tuned embeddings. We assess embeddings vulnerabilities across different types and dimensions, providing deeper insights into how task adaptation shifts privacy risks throughout genomic sequences. Our findings show a clear distinction in reconstruction vulnerability between pretrained and fine-tuned embeddings. Notably, fine-tuning strengthens resistance to reconstruction attacks in multiple architectures -- XLNet (+19.8\%), GPT-2 (+9.8\%), and BERT (+7.8\%) -- pointing to task-specific optimization as a potential privacy enhancement mechanism. These results highlight the need for advanced protective mechanisms for language models processing sensitive genomic data, while highlighting fine-tuning as a potential privacy-enhancing technique worth further exploration.

As large language models (LLMs) become progressively more embedded in clinical decision-support, documentation, and patient-information systems, ensuring their privacy and trustworthiness has emerged as an imperative challenge for the healthcare sector. Fine-tuning LLMs on sensitive electronic health record (EHR) data improves domain alignment but also raises the risk of exposing patient information through model behaviors. In this work-in-progress, we present an exploratory empirical study on membership inference vulnerabilities in clinical LLMs, focusing on whether adversaries can infer if specific patient records were used during model training. Using a state-of-the-art clinical question-answering model, Llemr, we evaluate both canonical loss-based attacks and a domain-motivated paraphrasing-based perturbation strategy that more realistically reflects clinical adversarial conditions. Our preliminary findings reveal limited but measurable membership leakage, suggesting that current clinical LLMs provide partial resistance yet remain susceptible to subtle privacy risks that could undermine trust in clinical AI adoption. These results motivate continued development of context-aware, domain-specific privacy evaluations and defenses such as differential privacy fine-tuning and paraphrase-aware training, to strengthen the security and trustworthiness of healthcare AI systems.

Large Language Models (LLMs) have achieved impressive performance across diverse natural language processing tasks, but their growing power also amplifies potential risks such as jailbreak attacks that circumvent built-in safety mechanisms. Existing defenses including input paraphrasing, multi step evaluation, and safety expert models often suffer from high computational costs, limited generalization, or rigid workflows that fail to detect subtle malicious intent embedded in complex contexts. Inspired by cognitive science findings on human decision making, we propose SafeBehavior, a novel hierarchical jailbreak defense mechanism that simulates the adaptive multistage reasoning process of humans. SafeBehavior decomposes safety evaluation into three stages: intention inference to detect obvious input risks, self introspection to assess generated responses and assign confidence based judgments, and self revision to adaptively rewrite uncertain outputs while preserving user intent and enforcing safety constraints. We extensively evaluate SafeBehavior against five representative jailbreak attack types including optimization based, contextual manipulation, and prompt based attacks and compare it with seven state of the art defense baselines. Experimental results show that SafeBehavior significantly improves robustness and adaptability across diverse threat scenarios, offering an efficient and human inspired approach to safeguarding LLMs against jailbreak attempts.

Instruction tuning has proven effective in enhancing Large Language Models' (LLMs) performance on downstream tasks. However, real-world fine-tuning faces inherent conflicts between model providers' intellectual property protection, clients' data privacy requirements, and tuning costs. While recent approaches like split learning and offsite tuning demonstrate promising architectures for privacy-preserving fine-tuning, there is a gap in systematically addressing the multidimensional trade-offs required for diverse real-world deployments. We propose several indicative evaluation metrics to guide design trade-offs for privacy-preserving fine-tuning and a series of example designs, collectively named GuardedTuning; they result from novel combinations of system architectures with adapted privacy-enhancement methods and emerging computation techniques. Each design represents distinct trade-offs across model utility, privacy guarantees, and costs. Experimental results demonstrate that these designs protect against data reconstruction attacks while maintaining competitive fine-tuning performance.