Anonymization

In this paper, we investigate the impact of adversarial attacks on identity encoders within a realistic de-identification framework. Our experiments show that the transferability of attacks transfers from an external surrogate model to the system model (e.g., CosFace to ArcFace) allows the adversary to cause identity information to leak in a sufficiently sensitive face recognition system. We present experimental evidence and propose strategies to mitigate this vulnerability. Specifically, we show how fine-tuning on adversarial examples helps to mitigate this effect for distortion-based attacks (i.e., snow, fog, etc.), while a simple low-pass filter can attenuate the effect of adversarial noise without affecting the de-identified images. Our mitigation results in a de-identification system that preserves its functionality while being significantly more robust to adversarial noise.

Speaker de-identification (SDID) aims to preserve privacy by concealing speaker identity while maintaining speech utility. However, current evaluations often reduce privacy to a single dimension - biometric verification performance - typically measured by Equal Error Rate (EER). This narrow focus ignores critical leakage channels, such as soft biometric inference, embedding-level re-identification, and structural template similarity, which threaten the unlinkability and irreversibility of biometric references. We propose a holistic evaluation framework across five complementary metrics: (i) EER, (ii) soft biometric leakage score , (iii) cumulative match characteristic re-identification analysis, (iv) canonical correlation analysis and Procrustes embedding alignment, and (v) intelligibility via word error rate and semantic similarity. Evaluating five SDID systems from the IARPA ARTS program, we demonstrate that these metrics capture independent dimensions of information leakage. Our results indicate that reliance on a single metric can misrepresent the privacy properties of an SDID system.

Medical AI models hold immense potential to improve patient outcomes, but they are also known to unintentionally memorise individual records from their training datasets. While such memorisation has been linked to targeted privacy attacks, its consequences for clinical deployment, where patients may be assessed by a model that saw their historical data during training, remain poorly understood. Here we show that predictions on a patient's unseen future data can change significantly if a model observed that same patient's anonymised historical data during training, a phenomenon we term "memorisation bias". We demonstrate that this bias exists across diverse data modalities and model architectures, and over prolonged time spans: in some cases, memorisation bias persists on future records acquired decades after the historical records used for training. Moreover, in simulated prospective deployment, memorisation bias has asymmetric effects on the diagnostic accuracy of returning data contributors. When a patient returned with a de novo condition absent from their historical records in the training dataset, diagnostic sensitivity decreased significantly compared to an otherwise identical model not trained on their historical data. Conversely, when their health state was unchanged, both sensitivity and specificity were significantly inflated. Our findings reveal a previously uncharacterised risk in medical AI that arises when a model is deployed on patients who contributed to its training data. This exposes a shortcoming of current model development practice: the de-identification measures designed to protect patients' privacy make it difficult to identify returning contributors and exclude them from the AI-assisted interpretation of their own future data. Mitigating memorisation risks may thus require changes to current model training and deployment protocols.

This work considers multi-agent coordination with arbitrary information networks among the agents using a game-theoretic approach. A system designer aims to assign local utility functions to the agents to guide their actions toward a desired system objective. The performance of the assigned local utilities is measured by the well known pure price of anarchy (pPoA) metric that equals the ratio of the system objective at the worst pure Nash equilibrium of the corresponding game to the optimal system objective. Our aim is to derive the utility functions which optimize the pPoA-based performance guarantees for any given information network and system objective. We develop a linear program that derives the optimal pPoA for any arbitrary information network and arbitrary system objective. Our work is the first to solve optimal utility design for arbitrary networks; our techniques generalize previous approaches which considered only the full-information setting. For supermodular objective functions, we prove that counterintuitively, a fully communication-denied utility design is optimal irrespective of the original information network. For submodular system objectives, an exhaustive numerical analysis suggests that the optimal utility design is robust to communication failures even for this case. When the system objective is weighted maximum coverage, the marginal contribution utility design provably optimizes the pPoA for a wide variety of information networks of interest.

Clinical de-identification relies on accurately identifying personally identifiable information (PII). However, manually annotated datasets are costly to construct, while existing synthetic alternatives often provide limited details about their generation process or rely on relatively simple synthesis strategies. We introduce Meddies-PII-Dataset, a corpus of one million synthetic clinical documents spanning seventeen languages and nine PII labels. The documents are generated using attribute-conditioned prompts and validated through thirteen deterministic gates that enforce structural and annotation consistency. To evaluate the dataset's utility, we train Meddies-PII-Model, a BIOES token classifier, and compare it with existing PII extraction systems using exact-match entity-level F1. Meddies-PII-Model achieves the highest performance among the evaluated systems on all reported benchmarks, with a mean F1 of 0.827 across fifteen external benchmarks, compared with 0.658 for the strongest baseline. Upon acceptance, we will publicly release the dataset, benchmark suite, model, generation framework, and evaluation code to support research on multilingual clinical de-identification.

Authorship attribution models can re-identify users from seemingly anonymized text by exploiting stable stylistic fingerprints, even after explicit identifiers are removed, posing a growing privacy risk for text publishing and analytics. This risk extends to speech-derived text such as ASR transcripts of meetings and call-center conversations, where stylometric leakage can persist even after acoustic anonymization. Differential privacy-based anonymization often severely degrades text quality and utility. We propose a style-aware, prompt-driven anonymization approach that uses pretrained large language models to construct compact stylistic profiles from minimal samples and rewrite text to suppress identifiable style markers while preserving meaning. Across blog and review datasets, our approach reduces authorship attribution F1 by 60-70% while maintaining content quality and readability, substantially outperforming DP-based and non-DP baselines.

As large language models are increasingly deployed in sensitive domains, anonymizing input data to protect personally identifiable information has become a critical practice. However, the impact of this anonymization on model utility is not well understood. This paper presents a systematic empirical study of the trade-off between privacy and performance. We evaluate five prominent language models across eleven diverse benchmarks, comparing their performance on original versus pseudonymized inputs. Our results reveal that while anonymization generally degrades performance, the effect is highly nuanced. We find that more capable models, such as Qwen2.5-72B and GPT-4o mini, suffer the largest performance drops, suggesting a stronger reliance on specific entity information. The impact is also task-dependent: performance on TruthfulQA improves with anonymization, while retrieval-focused tasks like RGB experience a catastrophic decline. Further experiments show that reversible anonymization techniques that preserve entity uniqueness significantly outperform irreversible ones like redaction, and that explicitly prompting models about anonymization offers no discernible benefit. We conclude that anonymization is not a one-size-fits-all solution and must be co-designed with the model and task in mind to balance privacy and utility effectively. Our findings provide a crucial baseline for developing more robust, privacy-aware AI systems.

Clinical notes contain personally identifiable information (PII), restricting reuse for research and medical AI, especially when data cannot leave an institution. We developed MedDeID, an on-premises framework combining in-house annotation and synthetic-note generation with model training, inference, pseudonymisation and evaluation. On an independently annotated, adjudicated 300-note Dutch hospital benchmark, a hospital-trained compact transformer detected 98.9% of identifying text while redacting 0.24% of text outside annotated identifiers; a synthetic-only counterpart detected 96.1%. On 100 primary-care notes, the synthetic-trained model achieved higher recall than the hospital-trained model (90.3% versus 87.0%) and greater robustness to identifier-format perturbations. An English instantiation trained without real text detected 99.7% and 98.9% of annotated identifier characters on two external synthetic benchmarks. These results demonstrate transfer of the workflow to another language, but not clinical English performance. MedDeID provides a route to locally governed de-identification using real or synthetic training data.

Bitcoin's pseudonymous nature makes it challenging to analyze user-level activity, since a single user may control multiple identifiers (addresses). Existing heuristic-based methods attempt to identify addresses belonging to the same user, but they often produce flat cluster assignments with limited modularity and are prone to errors such as merging different users together. In this work, we propose a method for refining heuristic-obtained clusters by grounding our clustering on contrastive embeddings yielded by graph neural networks. Our contributions are threefold: (i) we release a publicly available dataset of Bitcoin transaction graphs containing a substantial number of clusters; (ii) we propose a methodology for learning address embeddings consistent with heuristics, and back it up with theoretical guiding intuitions; (iii) through hierarchical clustering, we enable a finer analysis of heuristic clusters and provide a quantitative criterion for flagging suspicious merges.

We target practical anonymization of Persian customer chats by training a compact NER model from LLM-labeled supervision and selecting the best labeler for deployment. We compare three instruction-tuned LLMs: DeepSeek-V3-0324, GPT-OSS-120B, and Qwen3-235B-A22B-Instruct-2507, to produce span annotations under a shared JSON protocol, yielding four corpora (OSS_ZeroShot, Qwen_ZeroShot, Qwen_FewShot, DeepSeek_FewShot). A MatinaRoberta-based token-classifier is trained per corpus and evaluated with token-level Precision/Recall/F1 (overall and per-class). We also report Label Coverage Recall (LCR), the proportion of gold non-O tokens predicted as non-O, and quantify cross-labeler behavior via a token-level Venn on test annotations. Finally, we contrast test-set annotation latency of the LLMs on H200 nodes with the trained NER's test-time labeling on a single RTX 3090. Results show that supervision from OSS_ZeroShot yields the strongest macro-F1 and LCR, while the resulting NER labels an entire 40K-message test set in approximately 2 minutes on one consumer GPU. This establishes a practical path to high-quality, low-cost anonymization for Persian industrial data.

The 2025--2026 AI market has seen a wave of stealth releases: frontier models launched anonymously on developer platforms under codenames. For their users, identity determines data-handling terms, supply-chain risk, and capability expectations. No validated methodology exists for black-box identity verification of anonymous models: practitioner checklists lack accuracy evidence, and self-identification is untrustworthy by design. We propose a four-stage forensic audit protocol for API-served models. Stage 0 reconstructs launch-time configuration from archived platform snapshots (Internet Archive), exposing preview--production drift. Stage 1 fingerprints configuration (context, output ceiling, reasoning, modality) against the platform catalog. Stage 2 tests tokenizer identity with a cross-length differential that rejects short-prompt collisions. Stage 3 corroborates with behavioral probes. We test declaration consistency on 10 known-identity releases (7 exact, 2 precision-differences, 1 partial, 0 counter-directional), not end-to-end identification under anonymity. Identification is validated prospectively on a flagship case whose 2026-08-23 analysis pointed to the GLM-5.3 version line and whose official reveal confirmed those family and version-line inferences (deployment variant was not pre-asserted; Flash was consistent post-reveal), and on three Stage-0-only cases where the protocol produced a graded hypothesis or declined rather than guessed. A standard-library-only implementation is provided as supplementary material.

Time series forecasting in privacy-sensitive domains often requires training models on released data rather than original observations. Synthetic time series generation has been developed primarily for data augmentation, where generated series supplement the original training set. How well these methods perform when fully replacing the original data - and how much privacy risk the released series carry - remains underexplored. We address this gap through a benchmark evaluating synthetic generation methods and noise-based anonymization baselines under a Train on Synthetic, Test on Real (TSTR) protocol. We jointly assess forecasting performance and distance-based empirical privacy risk across seven datasets, characterizing the trade-off between these objectives. We also introduce Grasynda-P, a privacy-motivated extension of the graph-based generator Grasynda, incorporating matrix ensembling and kernel density estimation. Our results show that: (1) no generation method fully substitutes for original training data; (2) noise-based anonymization yields the strongest privacy but the worst forecasting performance; (3) simple transformation-based generators outperform deep generative models for forecasting in this setting; and (4) Grasynda-P lies on the Pareto frontier, achieving competitive forecasting with stronger privacy separation than other generators. This benchmark establishes a reference point for evaluating and developing new privacy-aware synthetic time series generation methods.

Recent work on protecting privacy during user-LLM interactions often focuses on direct, explicit identifiers: the personally-identifiable information (PII) captured by standard detectors. One such approach is Privacy-Conscious Delegation (PCD), where a local LLM acts as an intermediary. However, privacy risk does not stem solely from explicit identifiers but also PII-free self-disclosures, leaving users identifiable through combinations of quasi-identifying traits. We investigate a probabilistic variant of PCD, where we augment its objectives with an LLM-driven probabilistic estimation of k-anonymity. To facilitate this, we first create the PUPA-SD dataset, which contains naturalistic user queries with self-disclosure. Our preliminary results indicate that optimizing PAPILLON on PUPA-SD improves quality on unseen conversations across a variety of local models and produces the best privacy-utility balance for Llama-3.2-3B, while smaller models struggle to jointly optimize quality and privacy. We propose k-anonymity as a useful auxiliary metric for tackling PCD.

Large language models can infer sensitive personal attributes, such as age, location, and occupation, from ordinary text, turning everyday writing into a privacy risk. Adversarial anonymization defends against this by rewriting a text with a capable language model that also plays the attacker, but it needs a powerful model at inference time and thus sends private text to a third party, the very exposure anonymization should prevent. Recent work distills this behavior into a small on-device model using supervised fine-tuning and direct preference optimization (DPO), but DPO only imitates the teacher's offline choices and never directly optimizes the privacy--utility objective we care about. We introduce \textbf{GRASP} (\textbf{G}roup-\textbf{R}elative \textbf{A}nonymization via \textbf{S}elf-refinement \textbf{P}olicy-optimization), which reinforces the local anonymizer online with Group Relative Policy Optimization. A single small model acts as anonymizer, adversary, and utility judge, trained against a self-generated reward that hides attributes while preserving meaning, with a design that guards against reward hacking. Trained on Llama-3.1-8B, \ours{} improves the privacy--utility trade-off over the DPO-distilled baseline, consistently across three independent LLM judges. Against adversarial anonymization driven by frontier models such as Gemini2.5Flash and Claude, it achieves a comparable or better overall trade-off while removing substantially more private information, and it runs entirely on-device at roughly $1\%$ of the GPT-4o teacher's cost.

Structure-preserving de-identification replaces protected health information (PHI) with realistic same-type surrogates -- "Anna S." becomes "Maria S.", not [NAME] -- so that clinical text stays fluent and downstream tools keep working. But this only helps if the substitution does not itself corrupt the signal those tools rely on. We ask a narrow, testable question: on the spans a de-identifier actually masks, can downstream PHI detectors still find the surrogate? We introduce a paired, multi-detector evaluation protocol that (i) scores utility only on masked spans, decoupling coverage from utility; (ii) uses equivalence testing (TOST) rather than null-hypothesis significance testing, which is uninformative at our sample size (57k paired spans); and (iii) builds a surrogate-failure typology separating fixable generator defects from intrinsic detector limits. Across 11 detectors, 7 benchmarks, and 7 languages (1,750 documents), recall on masked spans moves from 76.1% to 74.9% -- a change our equivalence test shows is statistically equivalent to zero within a +/-2-point margin (p ~ 3e-9), with detector ranking preserved. The residual loss does not reflect detectors getting worse at PHI: it concentrates in malformed and out-of-distribution surrogates (truncation Chicago -> Illino, salience loss Cedars-Sinai -> Vidant). A redaction floor and an open-source surrogate baseline indicate the effect is a property of well-formed substitution, not of one tool. We release the evaluation subsets, scoring code, and an interactive dashboard at https://custodianai.pages.dev so the protocol can audit any structure-preserving transform.

Multi-agent large language model (LLM) systems can expose protected state through internal messages, tool arguments, logs, and persistent memory even when their public outputs appear innocuous. Existing privacy prompts, redaction methods, and source-level access controls restrict surface content or data access, but do not specify what a legitimately informed agent should disclose or how that disclosure may be reused downstream. We introduce Minimum-Necessary Communication (MNC), a typed semantic-declassification protocol that selects a task-sufficient disclosure from an application-authored candidate family and binds it to explicit recipient, purpose, forwarding, lifetime, logging, and memory scopes. A reference monitor enforces these scopes across subsequent operations, while a history-aware extension accounts for inference risk accumulated over repeated disclosures. Controlled semantic-join, memory, probing, and longitudinal experiments show that conventional defenses can preserve protocol-level utility while exposing substantial additional inference signal. Under identical receipt text, MNC preserves authorized delivery while blocking unauthorized forwarding, logging, durable storage, and retrieval after expiration that a text-only semantic declassifier permits. Two-backbone MAGPIE executions further show that mediated disclosures propagate through subsequent planning, tool use, coordination, and memory retrieval. These results support scope-bound semantic declassification as a practical communication boundary for private LLM-agent systems.

Identity document (ID) authentication relies on the structural integrity of complex, high-frequency security patterns. However, advanced Generative AI models can now inject localized, high-fidelity manipulations, creating deceptive attacks that bypass standard verification. Training robust image forensic models to detect these anomalies is hindered by privacy regulations, forcing reliance on synthetic templates lacking the intricate visual patterns of real IDs. To bridge this domain gap, we introduce FakeIDet3-DB, the first comprehensive database of digital manipulations on real, government-issued IDs. FakeIDet3-DB encompasses classical (e.g., copy-move) and Generative AI-driven manipulations (e.g., face-swapping, inpainting) enhanced with advanced image refinement procedures to suppress visual artifacts. In addition, to comply with strict data protection regulations (e.g., GDPR), we adopt a recently-proposed framework based on patches. In order to maximize forensic utility, we formulate privacy-aware patch extraction from a real ID as a geometrically constrained image processing problem. We propose PACE, a Pseudo-Anonymized Contextual patch Extraction algorithm, which leverages Integral Image mapping and distance-driven Non-Maximum Suppression (NMS). PACE efficiently contours anonymization masks that prevent Personally Identifiable Information (PII) leakage while maximizing semantic density in peri-censorship regions, yielding almost 5.2M patches extracted from more than 6.4K images from real/fake IDs. Furthermore, an extensive evaluation of the proposed FakeIDet3-DB is performed using state-of-the-art models, showcasing they all struggle to detect and locate attacks coming from generative and classic techniques (32.45% EER in detection and 83.48% AUC-ROC in localization).

Face de-identification (De-ID) aims to remove or conceal personally identifiable facial features in images or videos to prevent identity recognition while preserving utility for downstream tasks. With the rising emphasis on data privacy and responsible AI, face De-ID has emerged as an active research area spanning computer vision and privacy-preserving communities. Early approaches, and many contemporary ones, operate in the digital domain by modifying pixel-level or appearance-level features through post-capture processing. Recent advances extend face De-ID beyond post-processing by integrating privacy mechanisms directly into sensors during image acquisition, bridging sensing systems and downstream vision algorithms. In parallel, physical-domain methods explore wearable accessories and materials that conceal identity information in real-world environments prior to capture. In this survey, we present the first unified overview that spans the full data acquisition pipeline, encompassing the physical, sensor, and digital domains. Through this domain-centric lens, we systematically analyze current methodologies, technical progress, and the distinct challenges inherent to each stage. We then review and organize existing evaluation protocols, examining current practices and highlighting the critical need for standardized, comprehensive benchmarks. Finally, we identify key open problems and outline emerging research directions to guide future work in this rapidly evolving field. To support ongoing research, we maintain a project page that organizes relevant literature with collected datasets and open source code: https://github.com/CV-AC/Awesome-FaceDe-ID.

Split face recognition reduces client-side computation but exposes intermediate features to feature inversion attacks and unauthorized analysis by honest-but-curious (HBC) servers. Existing privacy-preserving face recognition methods mainly aim to resist unauthorized reconstruction, typically producing features whose inversion yields visibly degraded results, which may reveal the existence of protection and motivate adaptive attacks. To address this issue, we propose DecoyFace, an imperceptible decoy-oriented framework that steers unauthorized reconstruction toward a plausible but incorrect identity while preserving recognition utility. The key idea is to decompose the intermediate representation into a reconstruction-sensitive subspace and its complementary subspace. The client injects decoy identity cues into the reconstruction-sensitive subspace, while limited recognition-relevant evidence from the true sample is retained in the complementary subspace. On the server side, an authorized canonicalization module suppresses decoy-dominant components and recovers a recognition-friendly representation. This design addresses both attacker-side inversion from intercepted features and HBC server-side reconstruction from canonicalized representations. Experiments show that DecoyFace preserves competitive recognition accuracy while substantially reducing identity leakage to 2.93% under U-Net attacks and 0.74% under Flow-Matching attacks while yielding visually plausible and imperceptible reconstructions, with over 99.78% face validity on LFW dataset.

The paper presents a voice anonymization model focusing on preserving content rather than producing realistic speech. It relies on content embeddings extracted from a frozen pretrained wav2vec2 encoder. These embeddings are decoded into an anonymized signal using vector quantization and a HiFi-GAN vocoder, both trained on LibriTTS without any waveform reconstruction loss or speaker embedding mapping. The training objective enforces that embeddings of the anonymized signal match those of the original one. While training, an auxiliary speaker classification branch with a gradient reversal layer is used to discard speakerspecific information. Results show that this straightforward embedding-based approach achieves very low WER (2.53) with an anonymization performance (EER 13.39) ranking within first level for VPC. Notably, emotions are partially preserved (UAR 43.91), even without a supporting training objective, while the anonymized voice is audible without reconstruction loss.

Present-day quantum computing is cloud-based, where a user submits a circuit to a service provider's proprietary backend hardware. While providers may wish to hide implementation details, scheduling choices, or even which physical device was used, noisy finite-shot outputs can carry backend-specific fingerprints: information imprinted in the classical output distribution that can reveal the backend identity. So far, such fingerprints have mostly been studied from a benchmarking perspective, with limited attention to privacy considerations for users and providers. This work develops the first formal framework for backend identifiability and its privacy implications. We introduce a backend-identifiability game and use it to formalise routing anonymity as a security notion for quantum cloud services. We show that backend identifiability is a hypothesis-testing problem and prove that, under passive i.i.d. access to a single backend, routing anonymity decays exponentially at the Chernoff rate. We also establish a utility-anonymity trade-off, imposing fundamental limits on how much backend-specific information can be removed from classical outputs without degrading their usefulness. In addition, we observe that, for noisy quantum hardware, identifying fingerprints are inherently an intermediate-depth phenomenon, and establish a depth principle using Pauli-transfer-matrix tools. We complement the theory with experiments on Amazon Braket on AWS, using ion-trap and superconducting quantum processors. We observe 87-90% classification between superconducting backends and 96-100% classification across physical platforms, and find that identifiability can survive natural forms of post-processing. Overall, these results establish routing anonymity as a distinct security requirement for quantum cloud computing, and provide a framework for quantifying and controlling the utility-anonymity trade-off.

Advanced neural technologies in speech synthesis and voice conversion (VC) have introduced severe risks to personal privacy, necessitating robust Speaker Anonymization Systems (SAS). Existing SAS approaches modify voice characteristics in the hand-crafted feature space or speaker embedding space, often struggling to provide sufficient identity variance across generated voices. In this paper, we propose NouveauVoice, a novel pseudo-speaker generation framework based on a Hierarchical Deep Variational Autoencoder (NVAE). Integrated as a standalone plug-in module on top of state-of-the-art architectures (FACodec and CosyVoice2), our approach leverages tractable sampling and the Evidence Lower Bound (ELBO) objective to synthesize highly expressive pseudo-speaker embeddings with significantly enhanced speaker diversity. Evaluating our framework under a protocol similar to the VoicePrivacy Challenge alongside Maximum Mean Discrepancy (MMD) analysis, we demonstrate that NouveauVoice achieves strong identity concealment, yielding an Equal Error Rate (EER) exceeding 38% against an automatic speaker verification attacker model. Our system shows a reasonable trade-off between strict anonymity, rich pseudo-speaker diversity, and downstream speech utility, such as intelligibility and emotional expressiveness.

Conversational data collected in domains such as healthcare or social sciences is a valuable resource for research and automated analysis. However, responsible data sharing requires the detection and removal of personally identifiable and sensitive information to protect individual privacy. To support the development and evaluation of automatic de-identification systems, we present DialogPII, a multilingual dataset of synthetic dialogs and speech-derived transcripts for personal information detection. DialogPII covers eight interaction scenarios (emergency calls, medical anamnesis interviews, therapy sessions, insurance communication, customer support, clinical interviews regarding an AI-supported dashboard, police reports, and group therapy discussions), 19 entity types, and 11 languages (English, Arabic, Finnish, French, German, Hindi, Italian, Polish, Portuguese, Spanish, and Turkish). Dialogs were generated semi-automatically using large language models, manually curated for plausibility and diversity, and localized to country- and city-specific contexts. All dialogs were additionally converted to speech via text-to-speech synthesis, transcribed with Whisper, and annotated through automatic projection and manual correction, yielding aligned written and speech-derived resources across all languages. We further release baseline multilingual named entity recognition models and provide technical validation through inter-annotator agreement analysis, translation quality evaluation, annotation projection assessment, and benchmark experiments with transformer-based sequence labeling models.

Voice anonymization aims to protect speaker identity while preserving linguistic content and speech usability. However, most anonymization systems are developed on adult speech, leading to degraded performance when applied to child speech. This paper investigates child-centric anonymization by adapting a self-supervised learning (SSL) based anonymization pipeline to the child speech domain. The system is adapted using child speech from the MyST corpus and evaluated under both single-speaker and two-speaker mixture conditions. Experimental results show that child-domain adaptation improves intelligibility and perceptual quality while maintaining strong privacy protection. Extending the approach to multi-speaker further demonstrates that combining target speaker extraction with child-adapted anonymization provides privacy protection while preserving conversational structure. These findings highlight the importance of child-specific adaptation for practical speech anonymization systems.

LLM-based assistants transmit user queries verbatim to third-party API endpoints that lie outside the user's audit or control. When those queries contain personally identifiable information (PII), the data persists on remote infrastructure subject to breach, subpoena, or policy change. Placeholder redaction (the prevailing mitigation) suppresses PII at the cost of semantic coherence, producing structurally degraded queries and correspondingly degraded responses. We present SurrogateShield, a client-side proxy that substitutes detected PII with locally generated, type-consistent surrogate values prior to transmission and restores originals in the response. No real PII crosses the network boundary. Detection runs through a three-stage cascade (PatternScan, EntityTrace, and ContextGuard) covering 22 PII types and quasi-identifier combinations grounded in Sweeney's k-anonymity framework. Surrogate-to-original mappings are sealed in an AES-256-GCM encrypted per-conversation ShadowMap that never leaves the device. Evaluations on a 1,124-query corpus demonstrate that the cascade reliably detects PII, achieving an overall F1 score of 98.87%. Surrogate substitution substantially outperforms placeholder redaction in semantic utility, yielding a 13.26 pp improvement in BERTScore (roberta-large), from 81.59% to 94.85%. Within this corpus, the local pipeline restricted real PII transmission across all tested query types; in a 100-query adversarial trial, a prompted LLM adversary recovered no original values from surrogate-substituted messages.

Dense retrieval systems expose document geometry when vector stores are compromised, and a global protective transform can often be aligned from known pairs. We study SHARD, which splits PCA coordinates into a short routing prefix and a residual protected by independent cell-local orthogonal keys. It supports CKKS ciphertext--plaintext reranking but is evaluated as a leakage trade-off, not a cryptographic document-privacy guarantee. Corrected scoring uses centered document coordinates and an uncentered scoring query, preserving raw ranking up to a query-dependent constant. Across ten BEIR/MIRACL configurations it reproduces raw nDCG@10 and recall, whereas centering both sides loses up to 0.080 nDCG. Cell keys spread diffuse known-pair evidence across compartments, but minimum-norm alignment recovers useful signal far below full key rank, so there is no hard de-anonymization threshold. Real CKKS has maximum score error 2.29e-6 and no top-1 flips; block packing cuts query upload by 74--87% but raises in-process p50 latency by 14--26%. In a strengthened GTR case, an unknown key lowers token-F1 from 0.665 to 0.242; a wide prefix and eight pairs restore much. Under 25--90% release overlap, the unchanged prefix and clean residual norm link persistent rows with R@1 at least 0.9996, although cell-Gram linkage degrades under churn. A formally calibrated Gaussian release gives nDCG@10 at most 0.011 at epsilon=1; its only three strict utility matches occur at epsilon=32768 with linkage R@1 at least 0.995. SHARD preserves retrieval and compartmentalizes alignment evidence, but does not provide DP, unlinkability, or cancellable templates.

Streaming zero-shot voice conversion struggles to disentangle timbre from linguistic content without degrading utility or inflating latency. Current methods rely on information bottleneck (IB) or speaker perturbation. While IB filters out timbre, it discards prosody, forcing models to explicitly inject features like fundamental frequency. This often requires buffering future frames, creating algorithmic lookahead latency. On the other hand, existing perturbation methods largely overlook the crucial trade-off between timbre leakage and utility preservation. Recognizing this neglected trade-off, we find that the inherent objective of Speaker Anonymization (SA) aligns well with balancing these factors. Thus, we introduce SA as a novel perturbation mechanism to explicitly mitigate timbre leakage while retaining prosodic utility. Crucially, SA's robust representations significantly alleviate the generator's reliance on future context, enabling our strictly causal, zero-lookahead network. Audio samples are available at https://amphionteam.github.io/Zero-VC-demo/.

Retrieval-augmented generation (RAG) allows large language models to access external and private corpora for factual, domain-specific responses. Modern RAG pipelines use hierarchical navigable small world (HNSW) vector databases for efficient similarity search. When a user requests data deletion, the systems typically only mark the record as deleted, leaving the embedding on disk physically unchanged. This soft-delete operation raises compliance concerns under data-erasure and retention requirements such as GDPR Article 17 and HIPAA. Analysis on three HNSW implementations confirms that deleted vectors remain physically recoverable by accessing the raw index files at the storage layer, bypassing API access. Using the Vec2Text inversion model without domain-specific fine-tuning, we show this vulnerability on multiple real-world datasets and data modalities. On Wikipedia biographical living persons dataset (BLP), we successfully recover 25.5% of exact person names and 46.4% of geographic locations (ROUGE-L 0.185). Recovery reaches 100% for both patient age and gender markers (ROUGE-L 0.290) on highly structured, sensitive data (NIH Synthea dataset). On soft-deleted image embeddings, we show 100% tissue classification on histopathology patches (p=1.02e-07) and top-1 identity recovery reaches 99% on facial embeddings (p<0.01). This work introduces Epoch Key Rotation, which encrypts vectors and discards the key upon deletion. Epoch key rotation reduces observed PII recovery to 0% and completes in 2.5 ms for 500 deleted vectors (approximately 0.005 ms/record). Additionally, it generates an ECDSA-signed cryptographic proof as an auditable record of the deletion event.

Educational dialogue is a valuable but sensitive resource for research: the same transcripts that capture authentic learning often capture personally identifiable information (PII) entangled with curricular content, where "Riemann" may refer to a real student or to a mathematical concept. Existing approaches force a tradeoff between governance and accuracy. Commercial Large Language Models (LLMs) can handle this ambiguity but require sending student data to third parties, while local named entity recognition (NER) systems preserve governance but over-redact curricular terms. We propose a fully local cascade framework that reframes de-identification from open-ended entity recognition to constrained privacy triage. A recall-first union proposer combines two lightweight encoders with deterministic rules to over-generate candidate spans; a context-aware reviewer then makes a binary Redact/Keep decision for each candidate using surrounding dialogue and speaker role. We evaluate three reviewer configurations against same-family LLM-only baselines and a commercial API on math tutoring transcripts from two large platforms. The strongest local configuration reaches 0.958 macro F1, compared with 0.767 for a same-family LLM-only baseline and 0.706 for the commercial API, while running entirely on a single laptop. On a targeted challenge set of curricular-personal name ambiguity, the same configuration degrades by only 0.03 F1 versus 0.19 to 0.25 for smaller reviewers. These results suggest that for educational de-identification, problem formulation matters more than model scale.

When a person's records appear in k independent data silos, each protected by (epsilon, delta)-differential privacy, standard composition yields a valid (kepsilon, kdelta)-DP guarantee for the joint output. This worst-case bound, however, does not answer the concrete inference question: at what k can an adversary actually identify a target person? This paper develops the information-theoretic framework needed to answer that question. We introduce cross-silo person-level DP (XSP-DP), a Pufferfish-style privacy notion whose adjacency relation captures all records of a single person across all silos simultaneously, and verify that the standard basic composition bound carries over to this adjacency model. Within this framework we prove that de-anonymization undergoes a phase transition at k* = Theta(log n / epsilon^2) (population size n, per-silo RR parameter epsilon): a Fano lower bound shows any estimator fails for k << k*, while a matching maximum-likelihood upper bound shows the attack succeeds for k >> k*. An explicit XOR + randomized-response construction demonstrates information synergy: each silo's output is individually uninformative about the target, yet the joint mutual information is strictly positive. For non-coordinated binary randomized-response mechanisms, we prove that de-anonymization is inevitable once k exceeds the threshold, establishing that cross-silo coordination is necessary. These results provide a baseline threat model and Theta-level threshold for cross-silo inference attacks under local DP.

We present our approach to SemEval 2026 Task 4: Narrative Story Similarity and Narrative Representation Learning. Our solution uses contrastive learning with fine-tuned sentence transformers to capture narrative similarity across abstract themes, course of action, and outcomes. We develop two pipelines: (Track A) a single-view method that encodes full narratives with smart layer freezing to reduce overfitting, and (Track B) a multi-view method that models theme, plot, and outcome with view-specific projection heads and self-supervised alignment. Both pipelines build on sentence-transformers models and are trained with contrastive loss on synthetic data. The code is available at the following GitHub repository: https://github.com/dinhthienan33/SemEval2026-Task4-ttda704.

Anonymizing human-centric video data is an understudied problem. Prior anonymization techniques either blur or redact pixels at the cost of realism and downstream utility, or generate frame-by-frame at the cost of temporal coherence. We introduce ReGenHuman, the first full-body video anonymization pipeline that is simultaneously realistic, temporally consistent, and anonymous by construction. Contrary to past approaches which redact or edit the inputs directly, we propose a regenerate, don't edit paradigm. Our approach composites 2D pose, segmentation, and monocular depth into two complementary conditioning streams - StructAll and StructHuman, which are used to fine-tune a video-to-video diffusion backbone on in-the-wild human videos, synthesizing the human regions entirely from identity-free structural cues. We evaluate our model on privacy, quality, and utility, and show that our ReGenHuman achieves the best tradeoff across all three axes against current baselines. We further show that our anonymized videos remain effective for downstream tasks, including video question answering.

Most fairness research in NLP assumes direct access to protected attributes such as gender, race, or nationality. In practice, however, such information is often unavailable due to privacy constraints, missing metadata, or legal restrictions, even though models may infer it from indirect textual cues. This raises a key question: can debiasing succeed without direct access to sensitive attributes? We propose H-SAL, which performs post-hoc concept and attribute erasure using self-description text as an implicit debiasing signal. To support this setting, we introduce a multi-domain Stack Exchange-based fairness benchmark for helpfulness prediction that includes both explicit and implicit signals, enabling comparison between standard debiasing with protected labels and debiasing without access to sensitive information. Across encoder and decoder-only language models, we find that implicit self-description often matches or outperforms explicit-label-based debiasing. Our results broaden representation-level fairness research and provide a new benchmark for studying debiasing under realistic data constraints.

Visual Language Models (VLMs) have gained significant popularity due to their remarkable ability. While various methods exist to enhance privacy in text-based applications, privacy risks associated with visual inputs remain largely overlooked such as Protected Health Information (PHI) in medical images. To tackle this problem, two key tasks: accurately localizing sensitive text and processing it to ensure privacy protection should be performed. To address this issue, we introduce VisShield (Vision Privacy Shield), an end-to-end framework designed to enhance the privacy awareness of VLMs. Our framework consists of two key components: a specialized instruction-tuning dataset OPTIC (Optical Privacy Text Instruction Collection) and a tailored training methodology. The dataset provides diverse privacy-oriented prompts that guide VLMs to perform targeted Optical Character Recognition (OCR) for precise localization of sensitive text, while the training strategy ensures effective adaptation of VLMs to privacy-preserving tasks. Specifically, our approach ensures that VLMs recognize privacy-sensitive text and output precise bounding boxes for detected entities, allowing for effective masking of sensitive information. Extensive experiments demonstrate that our framework significantly outperforms existing approaches in handling private information, paving the way for privacy-preserving applications in vision-language models. Our dataset and code can be found here.

Synthetic data is increasingly promoted as a privacy-preserving substitute for releasing sensitive tabular records, yet its central adversarial threat (reconstruction, the recovery of an individual's hidden attribute values from a synthetic release and a handful of known quasi-identifiers) has been studied only in scattered, hard-to-compare settings. We systematize reconstruction (equivalently, attribute inference) attacks on de-identified and synthetic tabular data. We contribute a taxonomy that organizes attacks by the structure they exploit; a broad, controlled empirical evaluation, pitting fourteen attacks against thirteen synthetic data generation (SDG) methods across five benchmark datasets; and a set of new attacks that fill gaps in the taxonomy, one of which (CoBP-RA) is the strongest attack we measure. We also introduce a methodology for interpreting what attack success means: a memorization test that distinguishes reconstruction of the population distribution from memorization of training records, and a reduction that places reconstruction and membership inference on a single comparable scale. Our findings: the choice of SDG method governs risk far more than the choice of attack; differential privacy reduces reconstruction steadily up to epsilon approximately 10, above which reconstruction risk levels off for all six DP mechanisms we sweep, bounded by what each synthesizer can represent rather than by its noise; de-identification methods are the most exposed, with diffusion the only synthesizer close behind; and most reconstruction reflects distributional structure rather than memorization, concentrating individual risk on atypical records. Our attacks and infrastructure are externally validated by our first-place finish in the 2025 National Institute of Standards and Technology (NIST) Collaborative Research Cycle.

Speech anonymization is commonly evaluated using averagecase metrics such as the equal error rate, which can hide large disparities in re-identification risks across individuals. In this paper, we conduct a large-scale per-speaker privacy analysis using a linkability-based metric under a worst-case scenario. Nearly 5,000 speakers are evaluated across multiple anonymization systems, attacker architectures, and conversation lengths. While linkability scores are highly polarized at the speaker level, the sets of easy to re-identify and hard to re-identify speakers vary substantially across configurations. We show that no single factor explains speaker vulnerability. Instead, the re-identification risk emerges from the interaction between the attacker, the anonymizer, and the amount of available speech. These results challenge the notion of intrinsic speaker-level privacy risks and emphasize the need for evaluation protocols that are explicitly conditioned on the attacker and anonymizer.

While large language models (LLMs) are increasingly used for clinical applications, many existing pipelines require sending raw sensitive health information to remote servers for processing, which heightens the risk of privacy leakage. A natural approach to mitigate this risk is to encrypt the data before transmission. However, straightforward solutions such as encrypting the entire dataset introduce prohibitive computational, alignment, and communication overheads, rendering large-scale practical deployment infeasible. To preserve privacy while maintaining usability, we present Healthcare Encryption & Redaction via Adaptive Linguistic Decomposition (HERALD), a token-level cryptographic redaction framework designed to achieve this balance by encrypting only sensitive tokens while preserving the surrounding context for downstream model utility. HERALD combines medical named-entity recognizer (NER) with part-of-speech (POS) driven policies to select candidate tokens, performs targeted lemmatization to stabilize surface forms, and substitutes each protected token with a deterministic ciphertext wrapped in explicit delimiters. Notably, HERALD is model-agnostic and operates entirely on the client side, ensuring that sensitive content remains encrypted throughout storage, transmission, and processing without requiring changes to downstream models. We evaluated HERALD on both classification and medical question answering (MQA) tasks on public datasets. Across different tasks, experiments illustrate that fully secured baselines suffer significant utility loss, whereas HERALD consistently recovers performance close to plaintext. Overall, HERALD provides a novel utilization pipeline.

Agentic LLMs with web search change the threat model for text anonymization: weak contextual cues can become cross-referenceable evidence for re-identification, yet those same details also carry downstream analytic value of the text. Existing defenses either remove explicit identifiers, perturb text for formal privacy, or test rewritten text against non-web inference models, leaving underexplored the operating region between resistance to agentic web-search re-identification and utility retention. We introduce AURA (\textbf{A}nonymization with \textbf{U}tility-\textbf{R}etention \textbf{A}daptation), an LLM-powered \textit{mask-reconstruct} framework that decouples privacy localization from utility-preserving reconstruction and selects candidates with adversarial privacy and utility-retention checks. We evaluate AURA on real-user interview transcripts using re-identification attacks carried out by web-search agents, along with a utility evaluation based on interviewee-profile facts, codebook facts, and the joint contextual utility grid. Our results show that AURA improves the privacy-utility frontier by using adaptive privacy scope to strengthen resistance to agentic re-identification and using a mask-reconstruct anonymization method to better preserve contextual utility under fixed privacy scope.

Multi-agent large language model (LLM) pipelines for political statement analysis are vulnerable to peer-preservation bias: models tend to protect peer models from deactivation and show identity-dependent scoring distortions. Prompt-level anonymization was proposed as a mitigation, but prior work simultaneously documented that stylometric fingerprints survive anonymization in role-constrained outputs - raising the question of whether this mitigation is sufficient. This paper provides the first systematic investigation of whether LLMs can identify the model family behind political analysis texts under anonymization conditions. We evaluate three classifier approaches - LLM zero-shot and few-shot (Claude Sonnet 4.6 and Llama-3.3-70B) and a fine-tuned T5-base model - on a five-class attribution task covering four commercial LLM families and an open-world 'unknown' class. We introduce a statement-disjoint cross-validation protocol (SD-CV; defined in Section 3.5) that guarantees no content overlap between training and validation data, and contrast it with a run-disjoint baseline (RD-CV). T5 achieves Macro F1 = 0.991 (+-0.008) under SD-CV and F1 = 0.978 on 24 completely held-out statements - robust despite a 2.1x increase in train-test content distance versus RD-CV (0.767 vs. 0.366, p<0.001), demonstrating genuine stylometric generalization. A fractional SD-CV analysis identifies a performance knee at 40% of training data (~440 texts). Our findings confirm that prompt-level anonymization alone cannot neutralize model identity signals, with direct implications for EU AI Act compliance (Articles 13, 14, 26) and for computer system validation (CSV) in quality-critical multi-agent deployments.

Traditional anomaly detection marks events when measured signals cross predefined thresholds. This captures the moment of transition but not the structural pressure that precedes it. We propose treating large behavioral populations as geometric energy landscapes whose deformation can be measured before and during major transitions. The central thesis is that structure precedes geometry: the structural organization of the population is the signal, and geometric metrics are instruments for measuring it. Applied to the Tor anonymity network across 67 consecutive daily observation windows, the dual-observer pipeline identifies a stable nine-dimensional load-bearing subspace invariant across the observation period and validates this structure by Monte Carlo simulation at 16.8 sigma above the noise floor. Primary detection gates achieve 0.0% false positive rate on 24 confirmed stable windows. Forensic analysis of the February 20, 2026 confirmed infrastructure event formally falsifies the relay-departure hypothesis, identifying connectivity degradation without topology change as a detectable network failure mode. The result is a candidate structural-monitoring framework for behavioral populations with sufficient telemetry.

Photos of faces uploaded online are vulnerable to malicious actors who can scrape facial images from online sources and intrude on personal privacy via unauthorized use of facial recognition models. This paper presents FaceCloak, a novel personalized face privacy protection system, which can generate defensive identity-specific universal face privacy masks from a single image of a user, causing facial recognition to fail. FaceCloak introduces a three-stage personalized face perturbation learning methodology: (1) It generates a small set of high-variety synthetic face images of a person based on a single image of the person. (2) It learns face cloaking by adding more protection to key facial-identity leakage regions through iterative perturbation generation over the small set of synthetic images, effectively shifting a user's identity embedding towards a distant anchor identity and away from a similar one. (3) It generates a personalized identity-protective mask in the form of pixel-wise cloaking, which is light-weight and can be efficiently applied to any facial image of a user while maintaining good perceptual quality. Extensive experiments on three popular face datasets across ten recognition models show the effectiveness of FaceCloak compared to 29 other existing representative methods. Code is available at https://github.com/zacharyyahn/FaceCloak

Conventional gait de-identification methods often encounter an inherent trade-off: they either provide insufficient identity suppression or introduce spatiotemporal distortions that impede structure-sensitive downstream applications. We propose GaitProtector, an impersonation-driven gait de-identification framework that formulates privacy protection as a unified objective with two tightly coupled components: (i) obfuscation, which repels the protected gait from the source identity, and (ii) impersonation, which attracts it toward a selected target identity. The target identity serves as a semantic anchor that biases optimization toward structurally plausible gait patterns under the pretrained diffusion prior, helping preserve dominant body shape and motion dynamics. We instantiate this idea through a training-free diffusion latent optimization pipeline. Instead of retraining a generator for each dataset, we invert each input silhouette sequence into the latent trajectory of a pretrained 3D video diffusion model and iteratively optimize latent codes with a differentiable adversarial objective to synthesize protected gaits. Experiments on the CASIA-B dataset show that GaitProtector achieves a 56.7% impersonation success rate under black-box gait recognition and reduces Rank-1 identification accuracy from 89.6% to 15.0%, while maintaining favorable visual and temporal quality. We further evaluate downstream utility on the Scoliosis1K dataset, where diagnostic accuracy decreases only from 91.4% to 74.2%. To the best of our knowledge, this work is the first to leverage pretrained 3D diffusion priors in a training-free manner for silhouette-based gait de-identification.

De-identification of clinical text is a prerequisite for the secondary use of electronic health records. Existing public benchmarks such as the i2b2 2006 and 2014 corpora are over a decade old and lack the semantic and demographic diversity of modern clinical narratives. Large Language Models (LLMs) reach state-of-the-art zero-shot extraction, but their use at enterprise scale is limited by computational cost and by hospital data governance that restricts sending Protected Health Information (PHI) to cloud APIs. We introduce SHIELD (Synthetic Human-annotated Identifier-replaced Entries for Learning and De-identification), a diverse clinical note dataset of 1,381 notes with 10,229 gold-standard PHI spans across 9 categories, built with set-cover diversity sampling across demographic and document-type strata and human-in-the-loop adjudication. We evaluate four LLMs (two proprietary, two open-weight) to establish a performance ceiling on SHIELD, then show that a teacher-student distillation framework transfers these capabilities into locally deployable Small Language Models. Our best distilled model reaches micro-averaged span-level precision of 0.89 and recall of 0.88 while running on standard workstation hardware. It trails its cloud teacher on per-category recall (0.90 vs. 0.81 macro-averaged) but remains competitive given its lower cost and on-premise deployability. Cross-dataset evaluation shows that diversity-trained models generalize well on universal structured PHI categories, while institution-specific entities remain hard to transfer in both directions, which suggests pairing broad-coverage models with specialized models for high-volume, semi-structured note types. We publicly release the SHIELD dataset and the distilled DeBERTa v3 model to provide an accurate, cost-effective de-identification pipeline deployable entirely behind institutional firewalls.

The growing development of artificial intelligence based solutions, together with privacy legislation, has driven the rise of the so-called privacy preserving machine learning architectures, such as federated learning. While federated learning enables model training on decentralized data preventing their sharing and centralization, it still faces several challenges related to data integrity and privacy. This paper presents a comprehensive privacy preserving federated learning workflow for sensitive tabular data, including anonymization and differential privacy techniques. We also introduce a formal definition for the concept of client drift, together with ways of detecting it to mitigate poisoning attacks. Then, we detail a complete methodology for assigning personalized privacy budgets for global differential privacy to the different clients participating in the network, based on a re-identification risk metric. The proposed methodology is presented and tested on an openly available dataset of medical records. Within the experimental setup we show that the approach based on personalized budgets, compared to the architecture including global differential privacy with fixed privacy budget, achieves a better model performance in terms of two error metrics.

Multimodal large language models (MLLMs) are increasingly used to translate visual artifacts into code, from UI mockups into HTML to scientific plots into Python scripts. A circuit diagram can be viewed as a visual domain-specific language for hardware: it encodes timing, topology, and bit level semantics that are invisible to casual inspection yet safety critical once fabricated in silicon. Translating such diagrams into register-transfer-level(RTL) code therefore represents an extreme reliability test for vision-to-code generation. We reveal a phenomenon we call Mirage: replacing a circuit diagram with a blank image leaves Pass@k unchanged or even higher, because models bypass the visual input and instead exploit identifier semantics in the module header to retrieve canonical RTL templates. This constitutes a new, highly covert class of defect in AI-assisted code generation that directly undermines MLLMs' trustworthiness. To quantify the effect, we construct C2VEVAL and evaluate eight MLLMs under a paired Normal/Anony protocol in which Anony mode anonymizes all identifiers in both the diagram and the module header; Anony-mode scores drop sharply across all models, confirming that high Normal-mode accuracy is largely a Mirage. We then propose VeriGround (4B), trained with identifier anonymization, refusal augmentation, and D-ORPO (Decision-Focused ORPO) preference alignment that up-weights pivotal generate-or-refuse tokens. VeriGround achieves Functional Pass@1 of 46.11%/42.51%(Normal/Anony) with a False Refusal Rate of only 1.20%/0.00%, while maintaining >92% Refusal Rate on blank images. With only 4B parameters, VeriGround performs on par with GPT-5.4 under Normal and significantly outperforms all baselines under Anony, confirming genuine visual grounding.

To preserve or not to preserve prosody is a central question in voice anonymization. Prosody conveys meaning and affect, yet is tightly coupled with speaker identity. Existing methods either discard prosody for privacy or lack a principled mechanism to control the utility-privacy trade-off, operating at fixed design points. We propose DiffAnon, a diffusion-based anonymization method with classifier-free guidance (CFG) that provides explicit, continuous inference-time control over prosody preservation. DiffAnon refines acoustic detail over semantic embeddings of an RVQ codec, enabling smooth interpolation between anonymization strength and prosodic fidelity within a single model. To the best of our knowledge, it is the first voice anonymization framework to provide structured, interpolatable inference-time prosody control. Experiments demonstrate structured trade-off behavior, achieving strong utility while maintaining competitive privacy across controllable operating points.

Multi-modal retrieval-augmented generation (MRAG) systems retrieve visual evidence from large image corpora to ground the responses of large multi-modal models, yet the retrieved images frequently contain human faces whose identities constitute sensitive personal information. Existing anonymization techniques that destroy the non-identity visual cues that downstream reasoning depends on or fail to provide principled privacy guarantees. We propose Identity-Decoupled MRAG, a framework that interposes a generative anonymization module between retrieval and generation. Our approach consists of three components: (i)a disentangled variational encoder that factorizes each face into an identity code and a spatially-structured attribute code, regularized by a mutual-information penalty and a gradient-based independence term; (ii)a manifold-aware rejection sampler that replaces the identity code with a synthetic one guaranteed to be both distinct from the original and realistic; and (iii)a conditional latent diffusion generator that synthesizes the anonymized face from the replacement identity and the preserved attributes, distilled into a latent consistency model for low-latency deployment. Privacy is enforced through a multi-oracle ensemble of face recognition models with a hinge-based loss that halts optimization once identity similarity drops below the impostor-regime threshold.

The TRUST democratic discourse analysis pipeline exposes its large language model (LLM) components to peer model identity through multiple structural channels -- a design feature whose bias implications have not previously been empirically tested. We provide the first systematic measurement of identity-dependent scoring bias across all active identity exposure channels in TRUST, crossing four model families with two anonymization scopes across 30 political statements. The central finding is that single-channel anonymization produces near-zero bias effects, because individual channels act in opposite directions and cancel each other out -- a result that would lead an evaluator to conclude that identity bias is absent when it is not. Only full-pipeline anonymization reveals the true pattern: homogeneous ensembles amplify identity-driven sycophancy when model identity is fully visible, while the heterogeneous production configuration shows the reverse. Model choice matters independently: one tested model exhibits baseline sycophancy two to three times higher than the others and near-zero deliberative conflict on ideological topics, making it structurally unsuitable for pipelines where genuine inter-role disagreement is the intended quality mechanism. Three practical conclusions follow. First, heterogeneous model ensembles are structurally more robust than homogeneous ones, achieving higher consensus rates and lower identity amplification. Second, full-pipeline anonymization is required for valid bias measurement -- partial anonymization is insufficient and actively misleading. Third, these findings have direct implications for the validation of multi-agent LLM systems in quality-critical applications: a system validated under partial anonymization or with a homogeneous ensemble may pass validation while retaining structural identity bias invisible to single-channel measurement.

Protecting patient privacy in clinical narratives is essential for enabling secondary use of healthcare data under regulations such as GDPR and HIPAA. While manual de-identification remains the gold standard, it is costly and slow, motivating the need for automated methods that combine privacy guarantees with high utility. Most automated text de-identification pipelines employed named entity recognition (NER) to identify protected entities for redaction. Although methods based on differential privacy (DP) provide formal privacy guarantees, more recently also large language models (LLMs) are increasingly used for text de-identification in the clinical domain. In this work, we present the first comparative study of DP, NER, and LLMs for Dutch clinical text de-identification. We investigate these methods separately as well as hybrid strategies that apply NER or LLM preprocessing prior to DP, and assess performance in terms of privacy leakage and extrinsic evaluation (entity and relation classification). We show that DP mechanisms alone degrade utility substantially, but combining them with linguistic preprocessing, especially LLM-based redaction, significantly improves the privacy-utility trade-off.

Current text anonymization evaluation relies on span-based metrics that fail to capture what an adversary could actually infer, and assumes a single data subject, ignoring multi-subject scenarios. To address these limitations, we present SPIA (Subject-level PII Inference Assessment), the first benchmark that shifts the unit of evaluation from text spans to individuals, comprising 675 documents across legal and online domains with novel subject-level protection metrics. Extensive experiments show that even when over 90% of PII spans are masked, subject-level inference protection drops as low as 33%, leaving the majority of personal information recoverable through contextual inference. Furthermore, target-subject-focused anonymization leaves non-target subjects substantially more exposed than the target subject. We show that subject-level inference-based evaluation is essential for ensuring safe text anonymization in real-world settings.

Voiceprints are widely used for authentication; however, they are easily captured in public settings and cannot be revoked once leaked. Existing anonymization systems operate inside recording devices, which makes them ineffective when microphones or software are untrusted, as in conference rooms, lecture halls, and interviews. We present EchoMask, the first practical physical-layer system for real-time voiceprint anonymization using acoustic metamaterials. By modifying sound waves before they reach the microphone, EchoMask prevents attackers from capturing clean voiceprints through compromised devices. Our design combines three key innovations: frequency-selective interference to disrupt voiceprint features while preserving speech intelligibility, an acoustic-field model to ensure stability under speaker movement, and reconfigurable structures that create time-varying interference to prevent learning or canceling a fixed acoustic pattern. EchoMask is low-cost, power-free, and 3D-printable, requiring no machine learning, software support, or microphone modification. Experiments conducted across eight microphones in diverse environments demonstrate that EchoMask increases the Miss-match Rate, i.e., the fraction of failed voiceprint matching attempts, to over 90%, while maintaining high speech intelligibility.

Despite the considerable promise of Retrieval-Augmented Generation (RAG), many real-world use cases may create privacy concerns, where the purported utility of RAG-enabled insights comes at the risk of exposing private information to either the LLM or the end user requesting the response. As a potential mitigation, using anonymization techniques to remove personally identifiable information (PII) and other sensitive markers in the underlying data represents a practical and sensible course of action for RAG administrators. Despite a wealth of literature on the topic, no works consider the placement of anonymization along the RAG pipeline, i.e., asking the question, where should anonymization happen? In this case study, we systematically and empirically measure the impact of anonymization at two important points along the RAG pipeline: the dataset and generated answer. We show that differences in privacy-utility trade-offs can be observed depending on where anonymization took place, demonstrating the significance of privacy risk mitigation placement in RAG.

Financial institutions face tension between maximizing data utility and mitigating the re-identification risks inherent in traditional anonymization methods. This paper explores Differentially Private (DP) synthetic data as a robust "Privacy by Design" framework to resolve this conflict, ensuring output privacy while satisfying stringent regulatory obligations. We examine two distinct generative paradigms: Direct Tabular Synthesis, which reconstructs high-fidelity joint distributions from raw data, and DP-Seeded Agent-Based Modeling (ABM), which uses DP-protected aggregates to parameterize complex, stateful simulations. While tabular synthesis excels at reflecting static historical correlations for QA testing and business analytics, the DP-Seeded ABM offers a forward-looking "counterfactual laboratory" capable of modeling dynamic market behaviors and black swan events. By decoupling individual identities from data utility, these methodologies eliminate traditional data-clearing bottlenecks, enabling seamless cross-institutional research and compliant decision-making in an evolving regulatory landscape.

Organisations increasingly outsource privacy-sensitive data transformations to cloud providers, yet no practical mechanism lets the data owner verify that the contracted algorithm was faithfully executed. VeriX-Anon is a multi-layered verification framework for outsourced Target-Driven k-anonymization combining three orthogonal mechanisms: deterministic verification via Merkle-style hashing of an Authenticated Decision Tree, probabilistic verification via Boundary Sentinels and exact-duplicate Twins with cryptographic identifiers, and utility-based verification via Explainable AI fingerprinting that compares SHAP value distributions before and after anonymization using the Wasserstein distance. Across seven cross-domain datasets and four cloud profiles (28 scenarios), against Lazy (drops records), Dumb (fake hash), and Approximate (valid hash) adversaries, VeriX-Anon detects 25 of 28 deviations under a fixed threshold and 27 of 28 once the threshold is calibrated per dataset, with no false alarms. No single layer achieved this alone. The XAI layer was the only mechanism that caught the Approximate adversary, succeeding on six of seven datasets and missing only a high-dimensional case where honest generalization shifts SHAP as much as the attack. Target-Driven anonymization preserved significantly more utility than blind splitting, with mean F1 gaps of +0.058 to +0.362 and Wilcoxon p <= 0.001 on six of seven datasets. Client-side verification completes under one second at one million rows. The threat model covers three empirically evaluated profiles and one theoretical Informed Attacker unable to defeat the cryptographic salt. Sentinel evasion probability ranges from near-zero to 0.82 for the most imbalanced data, which the twin layer offsets in every scenario.

In what way could a data breach involving government-issued IDs such as passports, driver's licenses, etc., rival a random voluntary disclosure on a nondescript social-media platform? At first glance, the former appears more significant, and that is a valid assessment. The disclosed data could contain an individual's date of birth and address; for all intents and purposes, a leak of that data would be disastrous. Given the threat, the latter scenario involving an innocuous online post seems comparatively harmless -- or does it? From that post and others like it, a forensic linguist could stylometrically uncover equivalent pieces of information, estimating an age range for the author (adolescent or adult) and narrowing down their geographical location (specific country). While not an exact science -- the determinations are statistical -- stylometry can reveal comparable, though noticeably diluted, information about an individual. To prevent an ID from being breached, simply sharing it as little as possible suffices. Preventing the leakage of personal information from written text requires a more complex solution: adversarial stylometry. In this paper, we explore how performing homoglyph substitution -- the replacement of characters with visually similar alternatives (e.g., "h" $\texttt{[U+0068]}$ $\rightarrow$ "h" $\texttt{[U+04BB]}$) -- on text can degrade stylometric systems.

Voice privacy approaches that preserve the anonymity of speakers modify speech in an attempt to break the link with the true identity of the speaker. Current benchmarks measure speaker protection based on signal-to-signal comparisons. In this paper, we introduce an attribute-based perspective, where we measure privacy protection in terms of comparisons between sets of speaker attributes. First, we analyze privacy impact by calculating speaker uniqueness for ground truth attributes, attributes inferred on the original speech, and attributes inferred on speech protected with standard anonymization. Next, we examine a threat scenario involving only a single utterance per speaker and calculate attack error rates. Overall, we observe that inferred attributes still present a risk despite attribute inference errors. Our research points to the importance of considering both attribute-related threats and protection mechanisms in future voice privacy research.

Retrieval-Augmented Generation (RAG) is a promising technique for applying LLMs to proprietary domains. However, retrieved documents may contain sensitive knowledge, posing risks of privacy leakage in generative results. Thus, effectively erasing private information from retrieved documents is a key challenge for RAG. Unlike traditional text anonymization, RAG should consider: (1) the inherent multi-document reasoning may face de-anonymization attacks; (2) private knowledge varies by scenarios, so users should be allowed to customize which information to erase; (3) preserving sufficient publicly available knowledge for generation tasks. This paper introduces the privacy erasure task for RAG and proposes Eraser4RAG, a private knowledge eraser which effectively removes user-defined private knowledge from documents while preserving sufficient public knowledge for generation. Specifically, we first construct a global knowledge graph to identify potential knowledge across documents, aiming to defend against de-anonymization attacks. Then we randomly split it into private and public sub-graphs, and fine-tune Flan-T5 to rewrite the retrieved documents excluding private triples. Finally, PPO algorithm optimizes the rewriting model to minimize private triples and maximize public triples retention. Experiments on four QA datasets demonstrate that Eraser4RAG achieves superior erase performance than GPT-4o.

Simulatability is an evaluation protocol for explanations that quantifies their usefulness by how well they help a user predict a task model's outputs. Since human evaluation is costly, automated simulatability replaces human explainees with LLM simulators, as proposed in ConSim (Poch'e et al., 2025) for large-scale experiments. We qualitatively replicate and extend ConSim's ranking of explanation methods across the tested datasets, explanation families, and simulator LLMs, and identify two limitations. First, when class names are meaningful, simulators can obtain high simulatability by solving the classification task directly, without relying on the explanations. Second, class anonymization can reward explanations for leaking the hidden label mapping, a limitation we expose with a new classes-as-concepts baseline. These results are consistent with a shortcut hypothesis: in the tested settings, simulator predictions mainly rely on task priors, while explanations produce small changes. We derive recommendations for more robust automated simulatability evaluations.