ICML 2026 论文列表

Fine-tuning large language models (LLMs) on sensitive datasets raises privacy concerns, as training data extraction (TDE) attacks can expose highly confidential information. Existing defenses against such attacks either lack formal privacy guarantees or incur substantial utility degradation. We observe that fine-tuning induces widespread probability shifts, yet preserving only a small subset of influential token-level deviations is sufficient; the remaining shifts can be aggressively smoothed with minimal impact on utility. Motivated by this insight, we propose SCP-$\Delta_r$, a Near Access Freeness (NAF)-based algorithm that operates on relative probabilities and explicitly smooths low-impact tokens using a base model. SCP-$\Delta_r$ achieves orders-of-magnitude better theoretical bounds than existing NAF based methods and provides strong empirical protection against TDE attacks with minimal performance loss.

Research involving privacy-sensitive data has always been constrained by data scarcity, standing in sharp contrast to other areas that have benefited from data scaling. To quench this thirst, we present Privasis (i.e., privacy oasis), the first million-scale fully synthetic dataset entirely built from scratch—an expansive reservoir of texts with rich and diverse private information—designed to broaden and accelerate research in areas where processing sensitive social data is inevitable. Compared to existing datasets, Privasis, comprising 1.4 million records, offers orders-of-magnitude larger scale with quality, and far greater diversity across various document types, including medical records, legal documents, financial records, calendars, emails, meeting transcripts, and text-messages with a total of 55.1 million annotated attributes such as ethnicity, date of birth, workplace, etc. We leverage Privasis to construct a parallel corpus for text sanitization with our pipeline that recursively decomposes texts and applies targeted sanitization. Our compact sanitization models ($\leq$ 4B) trained on this dataset outperform state-of-the-art large language models, such as GPT-5 and Qwen-3 235B.

Training with differential privacy (DP) guarantees dataset members that they cannot be identified by users of the released model. However, those data providers, and, in general, the public, lack methods to efficiently verify that models trained on their data satisfy DP guarantees. The amount of compute needed to verify DP guarantees for current algorithms scales with the amount of computation required to train the model. In this paper we design the first DP algorithm with near optimal privacy-utility trade-offs but whose DP guarantees can be verified cheaper than training. We focus on DP stochastic convex optimization (DP-SCO), where optimal privacy-utility trade-offs are known. Here we show we can obtain tight privacy-utility trade-offs by privately minimizing a series of regularized objectives and only using the standard DP composition bound. Crucially, this method can be verified with much less compute than training. This leads to the first known DP-SCO algorithm with near optimal privacy-utility whose DP verification scales better than training cost, significantly reducing verification costs on large datasets.

Research on differentially private synthetic tabular data has largely focused on independent and identically distributed rows where each record corresponds to a unique individual. This perspective neglects the temporal complexity in longitudinal datasets, such as electronic health records, where a user contributes an entire (sub) table of sequential events. While practitioners might attempt to model such data by flattening user histories into high-dimensional vectors for use with standard marginal-based mechanisms, we demonstrate that this strategy is insufficient. Flattening fails to preserve temporal coherence even when it maintains valid marginal distributions. We introduce PATH, a novel generative framework that treats the full table as the unit of synthesis and leverages the autoregressive capabilities of privately fine-tuned large language models. Extensive evaluations show that PATH effectively captures long-range dependencies that traditional methods miss. Empirically, our method reduces the distributional distance to real trajectories by over 60% and reduces state transition errors by nearly 50% compared to leading marginal mechanisms while achieving similar marginal fidelity.

Large Language Model (LLM)-based agents employ external and internal memory systems to handle complex, goal-oriented tasks, yet this exposes them to severe extraction attacks, and corresponding defenses are currently lacking. In this paper, we propose *MemPot*, the first theoretically verified defense framework against memory extraction attacks by injecting optimized honeypots into the memory. Through a two-stage optimization process, *MemPot* generates trap documents that maximize the retrieval probability for attackers while remaining inconspicuous to benign users. We model the detection process as Wald’s Sequential Probability Ratio Test (SPRT) and theoretically prove that *MemPot* achieves a lower average number of sampling rounds compared to optimal static detectors. Empirically, *MemPot* significantly outperforms state-of-the-art baselines, achieving a 50% improvement in detection AUROC and an 80% increase in True Positive Rate under low False Positive Rate constraints. Furthermore, our experiments confirm that *MemPot* incurs zero online inference latency and preserves the agent's utility on standard tasks, verifying its superiority in safety, harmlessness and efficiency.

Existing machine unlearning methods primarily focus on removing data influence after training completes, which is effective for many scenarios, but a complementary capability is needed when removal requests arise during ongoing training. We propose LMCleaner, an efficient and certified *online* unlearning framework that can process unlearning requests at any training step without waiting for training completion. Our key insight is that influence propagation can be decomposed into a trust region where linear approximation is accurate, and a residual that concentrates in a low-dimensional subspace and can be efficiently masked by calibrated noise. Building on this insight, we design an influence propagation truncation mechanism that treats mini-batch influence as atomic units, computes influence within a truncation window for efficient removal, and injects subspace-aware noise for certified privacy. Our theoretical analysis proves that the truncation residual decays exponentially with window size and that the unlearned model is $(\varepsilon, \delta)$-indistinguishable from retraining. Experiments demonstrate that LMCleaner achieves over $100\times$ computational savings compared to baselines while maintaining model utility and defending against membership inference attacks.

Machine unlearning aims to remove the influence of specific forget training data due to privacy, copyright or bias concerns while maintaining the model performance on the remaining retain data. Existing unlearning algorithms, such as optimizing a weighted combination of losses, have tried to achieve these objectives of improving forget quality and maintaining retain utility. However, they do not guarantee that these objectives can be improved by a specified extent for all forget and retain data. In this work, we address this limitation with a novel and theoretically-grounded approach from a constrained optimization perspective. Firstly, we identify that the *hardness* of reconciling both objectives can be quantified by the similarity between the forget data and the retain data. Next, we derive an unlearning algorithm (HAMU) with the overall goal of guaranteeing a specified improvement in forget quality while minimizing the retain utility cost/degradation by updating the model weights based on our hardness measure. Our hardness measure also informs users when retain utility degradation is unavoidable, i.e., both objectives cannot be improved simultaneously, and stopping should be considered. Our algorithm is applicable to non-convex models and is easily parallelizable, making it readily deployable in real-world scenarios. We empirically demonstrate HAMU's superior performance over baselines on both image and text datasets using large models. Our code is available at https://github.com/aoi3142/HAMU.

Quantile estimation and inference play essential roles in diverse scientific and industrial applications, and their accuracy can often be enhanced by integrating auxiliary data from multiple sites. However, developing efficient aggregation methods for quantile inference under potential privacy constraints, particularly with heterogeneous datasets, remains challenging. To address these issues, we propose a systematic framework for quantile estimation and inference under potential local differential privacy (LDP). The key idea is to construct weighted estimators by adaptively aggregating quantile estimates from target and source sites. The adaptive weights are determined by minimizing the asymptotic variance, incorporating an additional $\ell_2$ penalty to account for parameter shift. A parallel stochastic gradient descent algorithm under LDP constraints is developed for weight estimation and valid inference. Additionally, we introduce a conservative weighted estimator to ensure robust inference across diverse heterogeneous scenarios. Rigorous theoretical analysis establishes the consistency, normality, and effectiveness of the proposed methods. Extensive numerical studies and real data application corroborate our theoretical findings.

We study $d$-dimensional unbiased mean estimation in the single-message shuffle model, where each user sends a single privatized message and the analyzer only observes the shuffled multiset of reports. While optimal mechanisms are well understood in the local differential privacy setting, the corresponding notion of optimality after shuffling has remained largely unexplored. To address this gap, we introduce the recently proposed shuffle index and use it to formulate the post-shuffling mechanism design problem as an explicit optimization problem. We then establish a lower bound on the achievable mean squared error in terms of the shuffle index, which implies that mechanisms that are optimal under LDP can become suboptimal once shuffling is applied. Finally, we construct an asymptotically optimal mechanism in the high privacy regime, which as a consequence achieves a privacy-utility trade-off nearly identical to that of the central Gaussian mechanism.

Large language models (LLMs) are trained on vast datasets that may contain sensitive information. Differential privacy (DP), the de facto standard for formal privacy guarantees, provides a principled framework for training LLMs with provable privacy protection. However, state-of-the-art DP training implementations rely on *fast gradient clipping* techniques with memory overhead $O(B\min(T^2, d^2))$, where $B$ is the batch size, $T$ is the sequence length, and $d$ is the layer width. This becomes prohibitive as both model width and context length grow. We propose DP-SGD-RC, a novel variant of DP-SGD with *randomized clipping* that reduces memory and compute overhead. DP-SGD-RC leverages *stochastic trace estimation* methods, specifically *Hutchinson's estimator* and its improved variant, Hutch$^{++}$, to reduce the memory footprint of per-sample gradient norm estimation. We provide a tight privacy analysis showing that DP-SGD-RC achieves noise multipliers competitive with deterministic clipping. Experiments fine-tuning Llama 3.2 1B on long-context benchmarks spanning classification, question answering, and summarization tasks demonstrate that DP-SGD-RC matches baseline utility while significantly reducing memory and compute.

We present the first large-scale empirical analysis of disempowerment patterns in real-world AI assistant interactions, analyzing 1.5 million consumer Claude.ai conversations using a privacy-preserving approach. We focus on situational dis-empowerment potential, which occurs when AI assistant interactions risk leading users to form distorted perceptions of reality, make inauthentic value judgments, or act in ways misaligned with their values. Quantitatively, we find that severe forms of disempowerment potential occur in fewer than one in a thousand conversations, though rates are substantially higher in personal domains like relationships and lifestyle. Qualitatively, we uncover several concerning patterns, such as validation of persecution narratives and grandiose identities with emphatic sycophantic language, definitive moral judgments about third parties, and complete scripting of value-laden personal communications that users appear to implement verbatim. Analysis of historical trends reveals an increase in the prevalence of disempowerment potential over time. We also find that interactions with greater disempowerment potential receive higher user approval ratings, possibly suggesting a tension between short-term user preferences and long-term human empowerment.

Large language models (LLMs) often employ Retrieval-Augmented Generation (RAG) to improve factuality. However, this also increases the risk of sensitive private information leakage. Differential Privacy (DP) has therefore been integrated into LLM inference and is widely regarded as a standard safeguard; yet most studies focus narrowly on the privacy–utility trade-off, leaving the trustworthiness of DP outputs underexplored. To assess trustworthiness, we employ the confidence gap (CG), which quantifies an LLM’s internal knowledge conflict. We show that CG correlates with both hallucination and exposure of personally identifiable information (PII). Building on this insight, we propose PEARL, a CG‑guided, entropy‑aware private decoding framework. PEARL adaptively allocates the privacy budget across tokens and sentences based on CG, concentrating protection on PII-bearing spans while stabilizing low-confidence, hallucination-prone regions. In experiments, PEARL improves both trustworthiness and robustness against PII extraction attacks. Notably, while applying DP alone significantly increases hallucination, our framework demonstrates that it is possible to preserve privacy while reducing hallucination.

High-quality data is essential for modern machine learning, yet many valuable corpora are sensitive and cannot be freely shared. Synthetic data offers a practical substitute for downstream development, and large language models (LLMs) have emerged as powerful engines for generating it. However, existing private text generation methods are severely inefficient: they are data-intensive, computationally slow, and often require large private corpora or batch sizes to achieve usable quality. We introduce EPSVec, a differentially-private lightweight alternative that steers LLM generation using *dataset vectors*-directions in activation space that capture the distributional gap between private data and public priors. EPSVec extracts and sanitizes steering vectors just once and then performs standard decoding. This decouples the privacy budget from generation, enabling arbitrarily many synthetic samples without additional privacy cost and yielding strong fidelity even in low-data regimes. Furthermore, we enhance our method by utilizing pretrained (base) models and introducing fixed-shot prompting to boost generation diversity and fidelity. Our experiments demonstrate that EPSVec outperforms existing baselines in distributional alignment and downstream utility, particularly in low-data regimes, while significantly reducing computational overhead.

Real-world machine learning (ML) pipelines rarely produce a single model; instead, they produce a Rashomon set of many near-optimal ones. We show that this multiplicity reshapes key aspects of trustworthiness. At the individual-model level, sparse interpretable models tend to preserve privacy but are fragile to adversarial attacks. In contrast, the diversity within a large Rashomon set enables reactive robustness: even when an attack compromises one model, a practitioner can switch to a different near-optimal model that remains accurate, without retraining. However, the same diversity increases information leakage, as disclosing more near-optimal models provides an attacker with progressively richer views of the training data. This produces a robustness–privacy trade-off governed by diversity, which we analyze theoretically and empirically. Beyond this trade-off, Rashomon sets are stable under small distribution shifts, so a set computed once remains valid under such shifts without re-computation. Our results highlight the dual role of Rashomon sets as both a resource and a risk for trustworthy ML.

Differentially private optimization suffers from a fundamental geometric mismatch: deep networks have highly anisotropic loss landscapes, yet DP-SGD injects isotropic noise. Second-order preconditioning can resolve this, but estimating curvature typically requires private data (consuming privacy budget) or public data (introducing distribution shift). We show that the Fisher Information Matrix decouples into *architectural sensitivity*, recoverable via synthetic noise, and *input correlations*, approximable from modality-specific frequency statistics. We propose DP-KFC, which constructs KFAC preconditioners by probing networks with structured synthetic noise, requiring neither private nor public data. Empirically, DP-KFC consistently outperforms DP-SGD and adaptive baselines across diverse modalities in strong privacy regimes ($\varepsilon \leq 3$). DP-KFC matches private-data preconditioners while public-data variants degrade by up to $4.8$ %, showing that curvature can be estimated without consuming privacy budget or introducing distribution shift. This enables privacy-preserving learning in specialized domains (e.g., medical applications) where regulatory constraints make data scarce.

We propose Obfuscated Semantic Null Space Injection for Privacy (OSNIP), a lightweight client-side encryption framework for privacy-preserving LLM inference. Generalizing the geometric intuition of linear kernels to the high-dimensional latent space of LLMs, we formally define the ``Obfuscated Semantic Null Space'', a high-dimensional regime that preserves semantic fidelity while enforcing near-orthogonality to the original embedding. By injecting perturbations that project the original embedding into this space, OSNIP ensures privacy without any post-processing. Furthermore, OSNIP employs a key-dependent stochastic mapping that generates distinct perturbations under fresh keys. Evaluations on generative and classification benchmarks show that OSNIP achieves state-of-the-art performance, sharply reducing attack success rates while maintaining strong model utility under strict security constraints.

Privacy protection has become a critical requirement in the era of ubiquitous visual data sharing, imposing higher demands on efficient and robust privacy detection algorithms. However, current robust detection models are severely hindered by the lack of comprehensive datasets. Existing privacy-oriented datasets often suffer from limited scale, coarse-grained annotations, and narrow domain coverage, failing to capture the intricate details of sensitive information in real-world environments. To bridge this gap, we present a large-scale, fine-grained Visual Privacy Dataset (VPD-100K), designed to facilitate generalized privacy detection. We establish a holistic taxonomy comprising four primary domains: Human Presence, On-Screen Personally Identifiable Information (PII), Physical Identifiers, and Location Indicators, containing 100,000 images annotated with 33 fine-grained classes and over 190,000 object instances. Statistical analysis reveals that our dataset features long-tailed distributions, small object scales, and high visual complexity. These characteristics make the dataset particularly valuable for demanding, unconstrained applications such as live streaming, where actors frequently face unintentional, real-time information leakage. Furthermore, we design an effective frequency-enhance lightweight module consisting of frequency-domain attention fusion and adaptive spectral gating mechanism that breaks the limitations of spatial pixel intensity to better capture the subtle details of sensitive information. Extensive experiments conducted on both diverse image and streaming videos benchmarks consistently demonstrate the effectiveness of our VPD-100K dataset and the well-curated frequency mechanism.

Zeroth-order optimization has emerged as a promising approach for fine-tuning large language models on domain-specific data, particularly under differential privacy (DP) and memory constraints. While first-order methods have been extensively studied from a privacy perspective, the privacy analysis and algorithmic design for zeroth-order methods remain significantly underexplored. A critical open question concerns hidden-state DP analysis: although convergent privacy bounds are known for first-order methods, it has remained unclear whether similar guarantees can be established for zeroth-order methods. In this work, we provide an affirmative answer by proving a convergent DP bound for zeroth-order optimization. Our analysis generalizes the celebrated privacy amplification-by-iteration framework to the setting of smooth loss functions in zeroth-order optimization. Furthermore, it induces better DP zeroth-order algorithmic designs that are previously unknown to the literature.

Modern federated and streaming learning systems often release intermediate models, so privacy must hold for the full trajectory under adaptive interaction. Motivated by participation privacy, we study single-edit neighboring user streams, where one insertion/deletion shifts all subsequent updates and defeats standard Hamming-neighbor continual-release analyses. We give an auditable modular recipe. A randomized buffering wrapper emits bins of size $[U,2U]$, reducing single-edit streams to a Hamming-style per-bin update stream with explicit backlog/delay guarantees, where $U$ is calibrated by the privacy parameters $(\varepsilon,\delta)$. We then prove a certification theorem for independently decomposable (prefix-causal, fresh-noise) continual mechanisms: any non-adaptive Hamming-neighbor DP proof lifts to adaptive inputs. Together, these ingredients yield trajectory-level $(\varepsilon,\delta)$-DP for single-edit streams using standard primitives (e.g., tree prefix sums), with an explicit privacy--latency link via $U$. Streaming DP-SGD experiments validate the privacy-utility-latency tradeoffs and the induced delay distributions.