Recent advances in deep generative models have made it increasingly difficult to distinguish real-world data from model-generated content. Consequently, successive model generations are trained on synthetic data from prior models, forming a self-consuming training loop. Despite being powerful, it also opens a new attack surface, i.e., test-time poisoning attacks, adversaries exploit model vulnerabilities during the inference stage to generate malicious test-time outputs that are subsequently incorporated into future training data. In this paper, for the first time, we propose a new class of malicious test-time poisoning attack against self-consuming generative methods. However, uncovering the failure modes of self-consuming generative models under test-time poisoning attacks is highly challenging due to the need to simultaneously maintain stealthy data manipulations while achieving fine-grained control over the activation and propagation of malicious behaviors across self-consuming training iterations. To address these challenges, our attacks provide fine-grained control and introduce a novel stealth paradigm that achieves both effectiveness and stealthiness by camouflaging stealthy malicious data within the training process and selectively activating concealed attack behaviors at targeted later stages. Notably, our attacks are compatible with various adversarial attack generating algorithms and self-consuming algorithms. Extensive experiments under different settings, including black-box scenarios, demonstrate that self-consuming generative models are broadly vulnerable to the proposed test-time poisoning attacks.

Large Reasoning Models (LRMs) have recently demonstrated significant improvements in complex reasoning. While quantifying generation uncertainty in LRMs is crucial, traditional methods are often insufficient because they do not provide finite-sample guarantees for reasoning-answer generation. Conformal prediction (CP) stands out as a model-agnostic methodology that constructs statistically rigorous uncertainty sets. However, existing CP methods ignore the logical connection between the reasoning trace and the final answer. Additionally, prior studies fail to interpret the origins of uncertainty coverage for LRMs as they typically overlook the specific training factors driving valid reasoning. Notably, it is challenging to disentangle reasoning quality from answer correctness, while simultaneously establishing theoretical guarantees for computationally efficient explanation methods. To address these challenges, we first propose a novel methodology that provides the uncertainty of the reasoning-answer structure with statistical guarantees. Subsequently, we develop a unified example-to-step explanation framework using Shapley values that identifies a provably sufficient subset of training data and their specific reasoning steps sufficient to achieve coverage. We also provide the theoretical analysis for our proposed methods. Extensive experiments on challenging reasoning datasets verify the effectiveness of the proposed methods.

Large language models (LLMs) have shown strong capabilities, enabling concise, context-aware answers in question answering (QA) tasks. The lack of transparency in complex LLMs has inspired extensive research aimed at developing methods to explain large language behaviors. Among existing explanation methods, natural language explanations stand out due to their ability to explain LLMs in a self-explanatory manner and enable the understanding of model behaviors even when the models are closed-source. However, despite these promising advancements, there is no existing work studying how to provide valid uncertainty guarantees for these generated natural language explanations. Such uncertainty quantification is critical in understanding the confidence behind these explanations. Notably, generating valid uncertainty estimates for natural language explanations is particularly challenging due to the auto-regressive generation process of LLMs and the presence of noise in medical inquiries. To bridge this gap, in this work, we first propose a novel uncertainty estimation framework for these generated natural language explanations, which provides valid uncertainty guarantees in a post-hoc and model-agnostic manner. Additionally, we also design a novel robust uncertainty estimation method that maintains valid uncertainty guarantees even under noise. Extensive experiments on QA tasks demonstrate the desired performance of our methods.