Large Vision-Language Models (LVLMs) are vulnerable to a growing array of multimodal jailbreak attacks, necessitating defenses that are both generalizable to novel threats and efficient for practical deployment. Many current strategies fall short, either targeting specific attack patterns, which limits generalization, or imposing high computational overhead. While lightweight anomaly-detection methods offer a promising direction, we find that their common one-class design tends to confuse unseen benign inputs with malicious ones, leading to unreliable over-rejection. To address this, we propose Representational Contrastive Scoring (RCS), a framework built on a key insight: the most potent safety signals reside within the LVLM’s own internal representations. Our approach inspects the internal geometry of these representations, learning a lightweight projection to maximally separate benign and malicious inputs in safety-critical layers. This enables a simple yet powerful contrastive score that differentiates true malicious intent from mere distribution shift. Our instantiations, MCD (Mahalanobis Contrastive Detection) and KCD (K-nearest Contrastive Detection), achieve state-of-the-art performance on a challenging evaluation protocol designed to test generalization to unseen attack types. This work demonstrates that effective jailbreak detection can be achieved by applying simple, interpretable statistical methods to the internal representations, offering a practical path towards safer LVLM deployment.

Knowledge distillation is a widely adopted technique for transferring capabilities from LLMs to smaller, more efficient student models.However, unauthorized use of knowledge distillation takes unfair advantage of the considerable effort and cost put into developing frontier models.We investigate methods for modifying teacher-generated reasoning traces to achieve two objectives that deter unauthorized distillation: (1) anti-distillation, or degrading the training usefulness of query responses, and (2) API watermarking, which embeds verifiable signatures in student models.We introduce several approaches for dynamically rewriting a teacher’s reasoning outputs while preserving answer correctness and semantic coherence.Two of these leverage the rewriting capabilities of LLMs, while others use gradient-based techniques.Our experiments show that a simple instruction-based rewriting approach achieves a strong anti-distillation effect while maintaining or even improving teacher performance.Furthermore, we show that our rewriting approach also enables embedding watermarks that can be reliably detectedwith essentially no false alarms.Our code is available at https://github.com/xhOwenMa/trace-rewriting.

To address data locality and privacy restrictions, Federated Learning (FL) has recently been adopted to fine-tune large language models (LLMs), enabling improved performance on various downstream tasks without requiring aggregated data. However, the repeated exchange of model updates in FL can result in prohibitively high communication costs, hindering the distributed learning process. To address this challenge, we propose EcoLoRA, a novel communication-efficient federated fine-tuning framework for LLMs. Leveraging the modular structure, we propose a round-robin segment sharing scheme, where each client uploads only a complementary LoRA segment per round to reduce network bandwidth. It is further combined with adaptive sparsification methods tailored to LoRA’s training dynamics and lossless encoding techniques. We conduct extensive evaluations on both question-answering and value-alignment tasks across multiple datasets and models. The results show that EcoLoRA significantly reduces communication overhead without compromising performance. For instance, it reduces communication time by up to 79% and total training time by up to 65%.

Prompt injection attacks pose a critical threat to large language models (LLMs), enabling goal hijacking and data leakage. Prompt guard models, though effective in defense, suffer from over-defense—falsely flagging benign inputs as malicious due to trigger word bias. To address this issue, we introduce NotInject, an evaluation dataset that systematically measures over-defense across various prompt guard models. NotInject contains 339 benign samples enriched with trigger words common in prompt injection attacks, enabling fine-grained evaluation. Our results show that state-of-the-art models suffer from over-defense issues, with accuracy dropping close to random guessing levels (60%). To mitigate this, we propose PIGuard, a novel prompt guard model that incorporates a new training strategy, Mitigating Over-defense for Free (MOF), which significantly reduces the bias on trigger words. PIGuard demonstrates state-of-the-art performance on diverse benchmarks including NotInject, surpassing the existing best model by 30.4%, offering a robust and open-source solution for detecting prompt injection attacks. The code and datasets are released at https://github.com/leolee99/PIGuard.