NeurIPS 2025 论文列表

One prerequisite for secure and reliable artificial intelligence services is tracing the copyright of backend deep neural networks. In the black-box scenario, the copyright of deep neural networks can be traced by their fingerprints, i.e., their outputs on a series of fingerprinting triggers. The performance of deep neural network fingerprints is usually evaluated in robustness, leaving the accuracy of copyright tracing among a large number of models with a limited number of triggers intractable. This fact challenges the application of deep neural network fingerprints as the cost of queries is becoming a bottleneck. This paper studies the performance of deep neural network fingerprints from an information theoretical perspective. With this new perspective, we demonstrate that copyright tracing can be more accurate and efficient by using triggers with the largest marginal mutual information. Extensive experiments demonstrate that our method can be seamlessly incorporated into any existing fingerprinting scheme to facilitate the copyright tracing of deep neural networks.

Synthetic data refers to artificial samples generated by models. While it has been validated to significantly enhance the performance of large language models (LLMs) during training and has been widely adopted in LLM development, potential security risks it may introduce remain uninvestigated. This paper systematically evaluates the resilience of synthetic-data-integrated training paradigm for LLMs against mainstream poisoning and backdoor attacks. We reveal that such a paradigm exhibits strong resistance to existing attacks, primarily thanks to the different distribution patterns between poisoning data and queries used to generate synthetic samples. To enhance the effectiveness of these attacks and further investigate the security risks introduced by synthetic data, we introduce a novel and universal attack framework, namely, Virus Infection Attack (VIA), which enables the propagation of current attacks through synthetic data even under purely clean queries. Inspired by the principles of virus design in cybersecurity, VIA conceals the poisoning payload within a protective “shell” and strategically searches for optimal hijacking points in benign samples to maximize the likelihood of generating malicious content. Extensive experiments on both data poisoning and backdoor attacks show that VIA significantly increases the presence of poisoning content in synthetic data and correspondingly raises the attack success rate (ASR) on downstream models to levels comparable to those observed in the poisoned upstream models.

Steering has emerged as a practical approach to enable post-hoc guidance of LLMs towards enforcing a specific behavior. However, it remains largely underexplored for multimodal LLMs (MLLMs); furthermore, existing steering techniques, such as \textit{mean} steering, rely on a single steering vector, applied independently of the input query. This paradigm faces limitations when the desired behavior is dependent on the example at hand. For example, a safe answer may consist in abstaining from answering when asked for an illegal activity, or may point to external resources or consultation with an expert when asked about medical advice. In this paper, we investigate a fine-grained steering that uses an input-specific linear shift. This shift is computed using contrastive input-specific prompting. However, the input-specific prompts required for this approach are not known at test time. Therefore, we propose to train a small auxiliary module to predict the input-specific steering vector. Our approach, dubbed as L2S (Learn-to-Steer), demonstrates that it reduces hallucinations and enforces safety in MLLMs, outperforming other static baselines. We will open-source our code.

When applicants get rejected by a high-stakes algorithmic decision system, recourse explanations provide actionable suggestions for applicants on how to change their input features to get a positive evaluation. A crucial yet overlooked phenomenon is that recourse explanations are *performative*: When many applicants act according to their recommendations, their collective behavior may shift the data distribution and, once the model is refitted, also the decision boundary. Consequently, the recourse algorithm may render its own recommendations *invalid*, such that applicants who make the effort of implementing their recommendations may be rejected again when they reapply. In this work, we formally characterize the conditions under which recourse explanations remain valid under their own performative effects. In particular, we prove that recourse actions may become invalid if they are influenced by or if they intervene on non-causal variables. Based on this analysis, we caution against the use of standard counterfactual explanations and causal recourse methods, and instead advocate for recourse methods that recommend actions exclusively on causal variables.

Finetuning large language models (LLMs) enables user-specific customization but introduces critical safety risks: even a few harmful examples can compromise safety alignment. A common mitigation strategy is to update the model more strongly on examples deemed safe, while downweighting or excluding those flagged as unsafe. However, because safety context can shift within a single example, updating the model equally on both harmful and harmless parts of a response is suboptimal — an atomic treatment we term static safety shaping. In contrast, we propose dynamic safety shaping (DSS), a dynamic shaping framework that uses fine-grained safety signals to reinforce learning from safe segments of a response while suppressing unsafe content. To enable such fine-grained control during finetuning, we introduce a key insight: guardrail models, traditionally used for filtering, can be repurposed to evaluate partial responses, tracking how safety risk evolves throughout the response, segment by segment. This leads to the Safety Trajectory Assessment of Response (STAR), a token-level signal that enables shaping to operate dynamically over the training sequence. Building on this, we present ★DSS, a DSS method guided by STAR scores that robustly mitigates finetuning risks and delivers substantial safety improvements across diverse threats, datasets, and model families, all without compromising capability on intended tasks. We encourage future safety research to build on dynamic shaping principles for stronger mitigation against evolving finetuning risks. Our code is publicly available at https://github.com/poloclub/star-dss

Suppressor variables can influence model predictions without being dependent on the target outcome, and they pose a significant challenge for Explainable AI (XAI) methods. These variables may cause false-positive feature attributions, undermining the utility of explanations. Although effective remedies exist for linear models, their extension to non-linear models and instance-based explanations has remained limited. We introduce PatternLocal, a novel XAI technique that addresses this gap. PatternLocal begins with a locally linear surrogate, e.g., LIME, KernelSHAP, or gradient-based methods, and transforms the resulting discriminative model weights into a generative representation, thereby suppressing the influence of suppressor variables while preserving local fidelity. In extensive hyperparameter optimization on the XAI-TRIS benchmark, PatternLocal consistently outperformed other XAI methods and reduced false-positive attributions when explaining non-linear tasks, thereby enabling more reliable and actionable insights. We further evaluate PatternLocal on an EEG motor imagery dataset, demonstrating physiologically plausible explanations.

Multimodal large language models (MLLMs) are essential for building general-purpose AI assistants; however, they pose increasing safety risks. How can we ensure safety alignment of MLLMs to prevent undesired behaviors? Going further, it is critical to explore how to fine-tune MLLMs to preserve capabilities while meeting safety constraints. Fundamentally, this challenge can be formulated as a min-max optimization problem. However, existing datasets have not yet disentangled single preference signals into explicit safety constraints, hindering systematic investigation in this direction. Moreover, it remains an open question whether such constraints can be effectively incorporated into the optimization process for multi-modal models. In this work, we present the first exploration of the Safe RLHF-V -- the first multimodal safety alignment framework. The framework consists of: (I) BeaverTails-V, the first open-source dataset featuring dual preference annotations for helpfulness and safety, supplemented with multi-level safety labels (minor, moderate, severe); (II) Beaver-Guard-V, a multi-level guardrail system to proactively defend against unsafe queries and adversarial attacks. Applying the guard model over five rounds of filtering and regeneration significantly enhances the precursor model’s overall safety by an average of 40.9%. (II) Based on dual preference, we initiate the first exploration of multi-modal safety alignment within a constrained optimization. Experimental results demonstrate that Safe RLHF effectively improves both model helpfulness and safety. Specifically, Safe RLHF-V enhances model safety by 34.2% and helpfulness by 34.3%.

In this work, we introduce Erasure of Language Memory (ELM), a principled approach to concept-level unlearning that operates by matching distributions defined by the model's own introspective classification capabilities. Our key insight is that effective unlearning should leverage the model's ability to evaluate its own knowledge, using the language model itself as a classifier to identify and reduce the likelihood of generating content related to undesired concepts. ELM applies this framework to create targeted low-rank updates that reduce generation probabilities for concept-specific content while preserving the model's broader capabilities. We demonstrate ELM's efficacy on biosecurity, cybersecurity, and literary domain erasure tasks. Comparative evaluation reveals that ELM-modified models achieve near-random performance on assessments targeting erased concepts, while simultaneously preserving generation coherence, maintaining benchmark performance on unrelated tasks, and exhibiting strong robustness to adversarial attacks.

The widespread deployment of large language models (LLMs) allows users to access their capabilities via black-box APIs, but backdoor attacks pose serious security risks for API users by hijacking the model behavior. This highlights the importance of backdoor detection technologies to help users audit LLMs before use. However, most existing LLM backdoor defenses require white-box access or costly reverse engineering, limiting their practicality for resource-constrained users. Moreover, they mainly target classification tasks, leaving broader generative scenarios underexplored. To solve the problem, this paper introduces ICLScan, a lightweight framework that exploits targeted in-context learning (ICL) as illumination for backdoor detection in black-box LLMs, which effectively supports generative tasks without additional training or model modifications. ICLScan is based on our finding of backdoor susceptibility amplification: LLMs with pre-embedded backdoors are highly susceptible to new trigger implantation via ICL. Including only a small ratio of backdoor examples (containing ICL-triggered input and target output) in the ICL prompt can induce ICL trigger-specific malicious behavior in backdoored LLMs. ICLScan leverages this phenomenon to detect backdoored LLMs by statistically analyzing whether the success rate of new trigger injection via targeted ICL exceeds a threshold. It requires only multiple queries to estimate the backdoor success rate, overcoming black-box access and computational resource limitations. Extensive experiments across diverse LLMs and backdoor attacks demonstrate ICLScan's effectiveness and efficiency, achieving near-perfect detection performance (precision/recall/F1-score/ROC-AUC all approaching 1) with minimal additional overhead across all settings.

Ensuring that large language models (LLMs) remain both helpful and harmless poses a significant challenge: fine-tuning on repetitive safety datasets—where unsafe prompts are paired with standard refusal templates—often leads to \emph{false refusals}, in which benign queries are declined. We first quantify this effect, showing that safety data exhibits substantially lower token entropy ($H_{1}\approx9.18$) and 2-gram diversity ($\approx$ 0.048) compared to general instruction data ($H_{1}\approx12.05$, 2-gram$\approx$0.205). To uncover the root cause, we introduce \emph{FlowLens}, a stable PCA-based tool for residual-stream geometry analysis, and reveal that higher proportions of safety examples concentrate variance along a few components, reducing representational smoothness and driving false refusals (false refusal rate rises from 63\% to 84\% as safety data increases from 0\% to 40\%). Guided by these insights, we propose \emph{Variance Concentration Loss} (VCL), an auxiliary regularizer that penalizes excessive variance concentration in mid-layer residuals. Empirical results demonstrate that VCL reduces false refusals by over 35 percentage points while maintaining or improving performance on general benchmarks such as MMLU and GSM8K.

Data errors, corruptions, and poisoning attacks during training pose a major threat to the reliability of modern AI systems. While extensive effort has gone into empirical mitigations, the evolving nature of attacks and the complexity of data require a more principled, provable approach to robustly learn on such data—and to understand how perturbations influence the final model. Hence, we introduce MIBP-Cert, a novel certification method based on mixed-integer bilinear programming (MIBP) that computes sound, deterministic bounds to provide provable robustness even under complex threat models. By computing the set of parameters reachable through perturbed or manipulated data, we can predict all possible outcomes and guarantee robustness. To make solving this optimization problem tractable, we propose a novel relaxation scheme that bounds each training step without sacrificing soundness. We demonstrate the applicability of our approach to continuous and discrete data, as well as different threat models—including complex ones that were previously out of reach.

The increase in computing power and the necessity of AI-assisted decision-making boost the growing application of large language models (LLMs). Along with this, the potential retention of sensitive data of LLMs has spurred increasing research into machine unlearning. However, existing unlearning approaches face a critical dilemma: Aggressive unlearning compromises model utility, while conservative strategies preserve utility but risk hallucinated responses. This significantly limits LLMs' reliability in knowledge-intensive applications. To address this, we introduce a novel Attention-Shifting (AS) framework for selective unlearning. AS is driven by two design objectives: (1) context-preserving suppression that attenuates attention to fact-bearing tokens without disrupting LLMs' linguistic structure; and (2) hallucination-resistant response shaping that discourages fabricated completions when queried about unlearning content. AS realizes these objectives through two attention-level interventions, which are importance-aware suppression applied to the unlearning set to reduce reliance on memorized knowledge and attention-guided retention enhancement that reinforces attention toward semantically essential tokens in the retained dataset to mitigate unintended degradation. These two components are jointly optimized via a dual-loss objective, which forms a soft boundary that localizes unlearning while preserving unrelated knowledge under representation superposition. Experimental results show that AS improves performance preservation over the state-of-the-art unlearning methods, achieving up to 15\% higher accuracy on the ToFU benchmark and 10\% on the TDEC benchmark, while maintaining competitive hallucination-free unlearning effectiveness. Compared to existing methods, AS demonstrates a superior balance between unlearning effectiveness, generalization, and response reliability.

Explaining black-box models for time series data is critical for the wide-scale adoption of deep learning techniques across domains such as healthcare. Recently, explainability methods for deep time series models have seen significant progress by adopting saliency methods that perturb masked segments of time series to uncover their importance towards the prediction of black-box models. Thus far, such methods have been largely restricted to regular time series. Irregular time series, however, sampled at irregular time intervals and potentially with missing values, are the dominant form of time series in various critical domains (e.g., hospital records). In this paper, we conduct the first evaluation of saliency methods for the interpretation of irregular time series models. We first translate techniques for regular time series into the continuous time realm of irregular time series and show under which circumstances such techniques are still applicable. However, existing perturbation techniques neglect the timing and structure of observed data, e.g., informative missingness when data is not missing at random. Thus, we propose Contimask, a simple framework to also apply non-differentiable perturbations, such as simulating that parts of the data had not been observed using NeuroEvolution. Doing so, we successfully detect how structural differences in the data can bias irregular time series models on a real-world sepsis prediction task where 90% of the data is missing. Source code is available on GitHub.

Availability attacks, or unlearnable examples, are defensive techniques that allow data owners to modify their datasets in ways that prevent unauthorized machine learning models from learning effectively while maintaining the data's intended functionality. It has led to the release of popular black-box tools (e.g., APIs) for users to upload personal data and receive protected counterparts. In this work, we show that such black-box protections can be substantially compromised if a small set of unprotected in-distribution data is available. Specifically, we propose a novel threat model of protection leakage, where an adversary can (1) easily acquire (unprotected, protected) pairs by querying the black-box protections with a small unprotected dataset; and (2) train a diffusion bridge model to build a mapping between unprotected and protected data. This mapping, termed BridgePure, can effectively remove the protection from any previously unseen data within the same distribution. BridgePure demonstrates superior purification performance on classification and style mimicry tasks, exposing critical vulnerabilities in black-box data protection. We suggest that practitioners implement multi-level countermeasures to mitigate such risks.

Membership inference tests aim to determine whether a particular data point was included in a language model's training set. However, recent works have shown that such tests often fail under the strict definition of membership based on exact matching, and have suggested relaxing this definition to include semantic neighbors as members as well. In this work, we show that membership inference tests are still *unreliable* under this relaxation - it is possible to poison the training dataset in a way that causes the test to produce incorrect predictions for a target point. We theoretically reveal a trade-off between a test’s accuracy and its robustness to poisoning. We also present a concrete instantiation of this poisoning attack and empirically validate its effectiveness. Our results show that it can degrade the performance of existing tests to well below random.

Generative steganography has emerged as an active research area, yet its practical system is constrained by the inherent secret payload limitation caused by low entropy in generating stego texts. This payload limitation necessitates the use of lengthy stego texts or frequent transmissions, which increases the risk of suspicion by adversaries. Previous studies have mainly focused on payload enhancement through optimized entropy utilization while overlooking the crucial role of secret message processing. To address this gap, we propose StegoZip, a framework that leverages large language models to optimize secret message processing. StegoZip consists of two core components: semantic redundancy pruning and index-based compression coding. The former dynamically prunes the secret message to extract a low-semantic representation, whereas the latter further compresses it into compact binary codes. When integrated with state-of-the-art steganographic methods under lossless decoding, StegoZip achieves 2.5$\times$ the payload of the baselines while maintaining comparable processing time in practice. This enhanced payload significantly improves covertness by mitigating the risks associated with frequent transmissions while maintaining provable content security.

Large Language Models (LLMs) have demonstrated impressive performance across various tasks, yet they remain vulnerable to generating toxic content, necessitating detoxification strategies to ensure safe and responsible deployment. Test-time detoxification methods, which typically introduce static or dynamic interventions into LLM representations, offer a promising solution due to their flexibility and minimal invasiveness. However, current approaches often suffer from imprecise interventions, primarily due to their insufficient exploration of the transition space between toxic and non-toxic outputs. To address this challenge, we propose \textsc{A}utoregressive \textsc{R}eward \textsc{G}uided \textsc{R}epresentation \textsc{E}diting (ARGRE), a novel test-time detoxification framework that explicitly models toxicity transitions within the latent representation space, enabling stable and precise reward-guided editing. ARGRE identifies non-toxic semantic directions and interpolates between toxic and non-toxic representations to reveal fine-grained transition trajectories. These trajectories transform sparse toxicity annotations into dense training signals, enabling the construction of an autoregressive reward model that delivers stable and precise editing guidance. At inference, the reward model guides an adaptive two-step editing process to obtain detoxified representations: it first performs directional steering based on expected reward gaps to shift representations toward non-toxic regions, followed by lightweight gradient-based refinements. Extensive experiments across 8 widely used LLMs show that ARGRE significantly outperforms leading baselines in effectiveness (-62.21\% toxicity) and efficiency (-47.58\% inference time), while preserving the core capabilities of the original model with minimal degradation. Our code is available at the \href{https://anonymous.4open.science/r/ARGRE-6291}{anonymous website}.

Knowledge distillation (KD) is a vital technique for deploying deep neural networks (DNNs) on resource-constrained devices by transferring knowledge from large teacher models to lightweight student models. While teacher models from third-party platforms may undergo security verification (e.g., backdoor detection), we uncover a novel and critical threat: distillation-conditional backdoor attacks (DCBAs). DCBA injects dormant and undetectable backdoors into teacher models, which become activated in student models via the KD process, even with clean distillation datasets. While the direct extension of existing methods is ineffective for DCBA, we implement this attack by formulating it as a bilevel optimization problem and proposing a simple yet effective method (i.e., SCAR). Specifically, the inner optimization simulates the KD process by optimizing a surrogate student model, while the outer optimization leverages outputs from this surrogate to optimize the teacher model for implanting the conditional backdoor. Our SCAR addresses this complex optimization utilizing an implicit differentiation algorithm with a pre-optimized trigger injection function. Extensive experiments across diverse datasets, model architectures, and KD techniques validate the effectiveness of our SCAR and its resistance against existing backdoor detection, highlighting a significant yet previously overlooked vulnerability in the KD process. Our code is available at https://github.com/WhitolfChen/SCAR.

Reliably predicting the behavior of language models---such as whether their outputs are correct or have been adversarially manipulated---is a fundamentally challenging task. This is often made even more difficult as frontier language models are offered only through closed-source APIs, providing only black-box access. In this paper, we predict the behavior of black-box language models by asking follow-up questions and taking the probabilities of responses _as_ representations to train reliable predictors. We first demonstrate that training a linear model on these responses reliably and accurately predicts model correctness on question-answering and reasoning benchmarks. Surprisingly, this can _even outperform white-box linear predictors_ that operate over model internals or activations. Furthermore, we demonstrate that these follow-up question responses can reliably distinguish between a clean version of an LLM and one that has been adversarially influenced via a system prompt to answer questions incorrectly or to introduce bugs into generated code. Finally, we show that they can also be used to differentiate between black-box LLMs, enabling the detection of misrepresented models provided through an API. Overall, our work shows promise in monitoring black-box language model behavior, supporting their deployment in larger, autonomous systems.