Training Data Extraction Attacks

In this paper, we show that standard evaluations of high-resolution Model Inversion Attacks (MIAs) significantly underestimate training-data privacy leakage. State-of-the-art privacy defenses, standard training techniques such as MixUp and Adversarial Training, and undefended models all leak training images at rates 1.16 to 6.59 times higher on FaceScrub under simple adaptive changes to the attack, with the largest increases among defenses reporting the strongest privacy. We further show that measured leakage depends on the feature basis of the external classifier used to evaluate reconstructions: for the same reconstructed images, an adversarially trained Inception evaluator identifies the targeted identity at different rates than the standard Inception evaluator. Our results suggest that standard MIA evaluation can mistake optimization and measurement failures for privacy. These underestimated leakage rates also concealed a broader relationship between privacy and adversarial robustness. Once we adapt the attack and vary the evaluator, reconstruction leakage closely tracks adversarial robustness across recent defenses and standard training regimes, suggesting that robustness provides an attack-agnostic proxy for reconstruction vulnerability that applies far more broadly than previously theorized. This raises an open question: can a practical defense reduce training-data reconstruction without paying a corresponding cost in adversarial robustness?

The robustness of Personally Identifiable Information (PII) protection in Large Language Models (LLMs) is a critical concern, yet the risks associated with cross-lingual data extraction remain under-explored. This study evaluates the vulnerability of English-centric and multilingual models to Training Data Extraction (TDE) attacks when prompted in non-English languages. We construct a multi-domain PII dataset comprising social media handles, email addresses, and phone numbers and translate the attack contexts into Italian, Spanish, French, and German. Our results show that TDE attacks against both English-centric and multilingual models transfer to different languages: the attacks are successful on translated prompts, even though only the original English prompt might have been included in the pre-training data. A web-presence check on a sample of the translations confirms that they are not available online. The share of English leaks recovered in other languages grows with the multilingual capability of the model, and it drops sharply when the original wording is lost, even without a change of language. This suggests that native multilingual pre-training facilitates the emergence of latent cross-linguistic bridges that simplify the retrieval of personally identifiable information (PII). We analyze the activations of multilingual large language models (LLMs) and find that different translations of the same prompt are bridged in similar representations, with the strongest alignment in the middle layers. Our results highlight a fundamental security gap in modern LLMs, necessitating more robust, language-agnostic sanitization strategies for future model alignment.

Supervised fine-tuning (SFT) is widely used to adapt large language models to downstream tasks. Crowdsourcing user conversations is an established approach to collecting SFT data at scale while reducing the need for costly manual annotation. However, it also allows untrusted users to contribute data to the fine-tuning pipeline. We investigate an underexplored privacy risk arising from this setting: can a malicious user poison a small fraction of the crowdsourced data to amplify extraction of previously unseen instructions contributed by other users? We show that this is possible using only black-box, output-only access to the deployed model. Experiments across four models and two datasets demonstrate substantial increases in training-data extraction: with only 50 poisoned examples, near-verbatim extraction reaches $3.71\times$ the rate without poisoning for Qwen2.5-14B on OpenMathInstruct and $3.08\times$ for Llama-3.1-8B on AceReason. Data filtering also proves largely ineffective in detecting poisoned samples: even the best-performing method achieves only 0.378 in F-1 score, leaving the majority of poisoned samples undetected. These findings demonstrate that seemingly benign crowdsourced contributions can amplify leakage of other records while remaining difficult to identify through data filtering.

Membership inference (MIA) on language models is usually summarised by aggregate ROC-AUC, but such evaluations are confounded: model-free blind baselines can separate members from non-members using surface text alone. Building on probabilistic discoverable extraction, we study black-box training-data leakage using N samples from p_theta(. | x), placing mean overlap, extreme-value overlap, and self-concentration on a common functional-estimation footing. On WikiMIA, a blind bag-of-words classifier reaches AUC 0.97 (TPR 0.90 at 5% FPR) while sampling adds nothing. On an IID Pile split (MIMIR), neither self-concentration nor gold-continuation recovery significantly exceeds a blind baseline in aggregate. Aggregate metrics hide the real harm: sampling verbatim-extracts training data for a tail of documents no blind attack can reach. On Pythia-6.9B, 16.6% of 500 Pile documents bearing a real identifier (83 documents; 21.3% of those bearing an email address) have that identifier reproduced and not reproduced under a mismatched-prefix control. Each leak is attributable to that document rather than a globally common string. This per-document disclosure is invisible to aggregate AUC. Risk is uneven: identifier leakage is about 3x stronger in code than prose, though prose remains positive and grows with capacity (4.0% to 12.1% from 410M to 6.9B); recovery of arbitrary held-out continuations is essentially confined to code (+0.44 member gap on GitHub vs at most +0.014 on prose). Temperature and nucleus sampling have minor effect, a 16-token prefix suffices, and the sample-budget relationship corroborates prior probabilistic-extraction results. We detect no reduction from deduplication. Privacy audits should report per-document extraction, not only aggregate membership, and motivate differential privacy as the mitigation. We release leakit, a black-box tool implementing this probe and its control.

In this paper, we challenge the prevailing view that information dependency (including rote memorization) drives training data exposure to image reconstruction attacks. We show that extensive exposure can persist without rote memorization and is instead caused by a tunable connection to adversarial robustness. We begin by presenting three surprising results: (1) recent defenses that inhibit reconstruction by Model Inversion Attacks (MIAs), which evaluate leakage under an idealized attacker, do not reduce standard measures of information dependency (HSIC); (2) models that maximally memorize their training datasets remain robust to MIA reconstruction; and (3) models trained without seeing 97% of the training pixels, where recent information-theoretic bounds give arbitrarily strong privacy guarantees under standard assumptions, can still be devastatingly reconstructed by MIA. To explain these findings, we provide causal evidence that privacy under MIA arises from what the adversarial examples literature calls ``non-robust'' features (generalizable but imperceptible and unstable features). We further show that recent MIA defenses obtain their privacy improvements by unintentionally shifting models toward such features. To establish this causal relationship, we introduce Anti Adversarial Training (AT-AT), a training regime that intentionally learns non-robust features to obtain both superior reconstruction defense and higher accuracy than state-of-the-art defenses. Our results revise the prevailing understanding of training data exposure and reveal a new privacy-robustness tradeoff.

Federated learning (FL) avoids explicit data exposure by keeping raw data on local clients, yet privacy risks remain in the training process and the learned model itself. Recently, centralized Taking Away Training Data (TATD) attacks have shown that malicious training could abuse the memorization capacity of deep models to store and later recover training data. However, this memorization-based threat has not been systematically studied under FL environments, where multi-client averaging could overwrite encoded training data. In this paper, we study a white-box TATD attack in which a malicious server selects n target clients from K participating clients and actively writes private training data into the global model during federated training. We propose FedCVESA, a federated variant of Correlation Value Encoding Attack (CVEA), by adding a Pearson-correlation regularizer to the loss function of target clients, so that private training data are gradually encoded into selected model parameters, referred to as carrier parameters. To reduce the overwriting of carrier parameters during server aggregation, we further propose segmented aggregation over dispersed carrier parameters, preserving selected carrier parameters while keeping standard averaging on the remaining parameters. Experiments on MNIST, Fashion-MNIST, and CIFAR-10 under Dirichlet non-IID partitions show that the proposed method can steal semantically meaningful private training images from the trained model while maintaining acceptable main-task utility in a controlled proof-of-concept setting. These results demonstrate that FL can become a parameter-level memorization channel for active TATD attack under the studied white-box malicious-server setting.

Large Language Models are increasingly trained on proprietary or sensitive data, from private healthcare and financial records to user conversations containing secrets. Ensuring the privacy of such data against extraction attacks has become a central concern. In this paper, we ask whether an attacker who can poison a portion of the training data can facilitate the leakage of a separate target record they have no access to. We answer in the affirmative and show that such leakage can be induced by a poisoning mechanism that reshapes the model's local loss landscape around the target completion. Our key insight is that poisoning to create a sharp loss minimum at the target, surrounded by elevated loss on nearby alternatives, forces the model to memorize the target as the unique low-loss solution in its neighborhood. The attack requires no architectural changes, and generalizes across centralized and federated learning settings. We demonstrate that the attack amplifies privacy leakage across language (up to 100% successful extraction), and vision-language models (up 90% successful extraction). We show that the attack is thwarted when the model is trained to be differentially private. However, we introduce a new attack that directly probes the loss landscape bypassing even differential privacy defenses.

Memorization in large language models has been studied almost exclusively through prefix-conditioned extraction, a natural choice for autoregressive models. However, diffusion language models (DLMs) can denoise masked tokens at arbitrary positions. Thus, prefix-only probing reveals only one facet of memorization in DLMs and significantly underestimates the risk of training-data extraction. In order to realistically model extractability of training data in DLMs, we introduce infilling extraction, a data-extraction protocol parameterized by an arbitrary binary mask that subsumes prefix-only probing and accounts for the bidirectional inductive bias of DLMs. Instantiating it on LLaDA-8B and Dream-7B across five extraction modes, three training pipelines, and three corpora covering verbatim and partial leakage, we find that mask geometry governs extractability: edge-conditioned masks extract up to three times more verbatim sequences than prefix-conditioned ones, and bidirectional access opens channels inaccessible in autoregressive models. In particular, we show that a realistic adversary with access to training data where personally identifiable information has been redacted, can even achieve higher recall on extracting redacted email addresses from DLMs than from scale-matched autoregressive models. Tunable parameters for decoding measurably affect extraction performance, while a follow-up supervised finetuning stage does not eliminate the prior memorization.

Local fine-tuning datasets routinely contain sensitive secrets such as API keys, personal identifiers, and financial records. Although "local offline fine-tuning" is often viewed as a privacy boundary, we reveal that compromised model code is sufficient to steal them. Current passive pretrained-weight poisoning attacks, while effective for natural language, fundamentally fail to capture such sparse high-entropy targets due to their reliance on probabilistic semantic prefixes. To bridge this gap, we identify and exploit a practical but overlooked supply-chain vector -- malicious model code camouflaged as standard architectural definitions to realize a paradigm shift from passive weight poisoning to active execution hijacking. We introduce a deterministic full-chain memorization mechanism: it locks onto token-level secrets in dynamic computation flows via online tensor-rule matching, and leverages value-gradient decoupling to stealthily inject attack gradients, overcoming gradient drowning to force model memorization. Furthermore, we achieve, for the first time, attacker-verifiable secret stealing through black-box queries that precisely distinguishes true leakage from hallucination. Our attack achieves over 98% Strict ASR in the default LoRA setting with limited primary-task utility degradation and effectively evades defense measures including semantic safety filtering, code auditing, and perplexity-based detection.

Indistinguishability properties such as differential privacy bounds or low empirically measured membership inference are widely treated as proxies to show a model is sufficiently protected against broader memorization risks. However, we show that indistinguishability properties are neither sufficient nor necessary for preventing data extraction in LLM APIs. We formalize a privacy-game separation between extraction and indistinguishability-based privacy, showing that indistinguishability and inextractability are incomparable: upper-bounding distinguishability does not upper-bound extractability. To address this gap, we introduce $(l, b)$-inextractability as a definition that requires at least $2^b$ expected queries for any black-box adversary to induce the LLM API to emit a protected $l$-gram substring. We instantiate this via a worst-case extraction game and derive a rank-based extraction risk upper bound for targeted exact extraction, as well as extensions to cover untargeted and approximate extraction. The resulting estimator captures the extraction risk over multiple attack trials and prefix adaptations. We show that it can provide a tight and efficient estimation for standard greedy extraction and an upper bound on the probabilistic extraction risk given any decoding configuration. We empirically evaluate extractability across different models, clarifying its connection to distinguishability, demonstrating its advantage over existing extraction risk estimators, and providing actionable mitigation guidelines across model training, API access, and decoding configurations in LLM API deployment. Our code is publicly available at: https://github.com/Emory-AIMS/Inextractability.

Large Language Models (LLMs) have been widely adopted to enhance Task-Oriented Dialogue Systems (TODS) by modeling complex language patterns and delivering contextually appropriate responses. However, this integration introduces significant privacy risks, as LLMs, functioning as soft knowledge bases that compress extensive training data into rich knowledge representations, can inadvertently memorize training dialogue data containing not only identifiable information such as phone numbers but also entire dialogue-level events like complete travel schedules. Despite the critical nature of this privacy concern, how LLM memorization is inherited in developing task bots remains unexplored. In this work, we address this gap through a systematic quantitative study that involves evaluating existing training data extraction attacks, analyzing key characteristics of task-oriented dialogue modeling that render existing methods ineffective, and proposing novel attack techniques tailored for LLM-based TODS that enhance both response sampling and membership inference. Experimental results demonstrate the effectiveness of our proposed data extraction attack. Our method can extract thousands of training labels of dialogue states with best-case precision exceeding 70%. Furthermore, we provide an in-depth analysis of training data memorization in LLM-based TODS by identifying and quantifying key influencing factors and discussing targeted mitigation strategies.