Loss Landscape Poisoning: Targeted Extraction of Unseen Training Data from LLMs

• URL: http://arxiv.org/abs/2606.17110v2
• Date: Mon, 22 Jun 2026 18:00:00 GMT
• Title: Loss Landscape Poisoning: Targeted Extraction of Unseen Training Data from LLMs
• Abstract summary: Large Language Models are increasingly trained on proprietary or sensitive data.<n>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.<n>We show that such leakage can be induced by a poisoning mechanism that reshapes the model's local loss landscape.
• Score: 5.474954219332285
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: 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.

Related papers

• Unlearned but Not Forgotten: Data Extraction after Exact Unlearning in LLM [31.093224824043087]
  We introduce a novel data extraction attack that leverages signals from the pre-unlearning model to guide the post-unlearning model.<n>We demonstrate our attack's effectiveness on a simulated medical diagnosis dataset to highlight real-world privacy risks associated with exact unlearning.
  arXiv  Detail & Related papers  (2025-05-30T09:09:33Z)
• Venomancer: Towards Imperceptible and Target-on-Demand Backdoor Attacks in Federated Learning [16.04315589280155]
  We propose Venomancer, an effective backdoor attack that is imperceptible and allows target-on-demand. The method is robust against state-of-the-art defenses such as Norm Clipping, Weak DP, Krum, Multi-Krum, RLR, FedRAD, Deepsight, and RFLBAT.
  arXiv  Detail & Related papers  (2024-07-03T14:22:51Z)
• No Vandalism: Privacy-Preserving and Byzantine-Robust Federated Learning [18.1129191782913]
  Federated learning allows several clients to train one machine learning model jointly without sharing private data, providing privacy protection. Traditional federated learning is vulnerable to poisoning attacks, which can not only decrease the model performance, but also implant malicious backdoors. In this paper, we aim to build a privacy-preserving and Byzantine-robust federated learning scheme to provide an environment with no vandalism (NoV) against attacks from malicious participants.
  arXiv  Detail & Related papers  (2024-06-03T07:59:10Z)
• Can Sensitive Information Be Deleted From LLMs? Objectives for Defending Against Extraction Attacks [73.53327403684676]
  We propose an attack-and-defense framework for studying the task of deleting sensitive information directly from model weights. We study direct edits to model weights because this approach should guarantee that particular deleted information is never extracted by future prompt attacks. We show that even state-of-the-art model editing methods such as ROME struggle to truly delete factual information from models like GPT-J, as our whitebox and blackbox attacks can recover "deleted" information from an edited model 38% of the time.
  arXiv  Detail & Related papers  (2023-09-29T17:12:43Z)
• Avoid Adversarial Adaption in Federated Learning by Multi-Metric Investigations [55.2480439325792]
  Federated Learning (FL) facilitates decentralized machine learning model training, preserving data privacy, lowering communication costs, and boosting model performance through diversified data sources. FL faces vulnerabilities such as poisoning attacks, undermining model integrity with both untargeted performance degradation and targeted backdoor attacks. We define a new notion of strong adaptive adversaries, capable of adapting to multiple objectives simultaneously. MESAS is the first defense robust against strong adaptive adversaries, effective in real-world data scenarios, with an average overhead of just 24.37 seconds.
  arXiv  Detail & Related papers  (2023-06-06T11:44:42Z)
• RelaxLoss: Defending Membership Inference Attacks without Losing Utility [68.48117818874155]
  We propose a novel training framework based on a relaxed loss with a more achievable learning target. RelaxLoss is applicable to any classification model with added benefits of easy implementation and negligible overhead. Our approach consistently outperforms state-of-the-art defense mechanisms in terms of resilience against MIAs.
  arXiv  Detail & Related papers  (2022-07-12T19:34:47Z)
• Truth Serum: Poisoning Machine Learning Models to Reveal Their Secrets [53.866927712193416]
  We show that an adversary who can poison a training dataset can cause models trained on this dataset to leak private details belonging to other parties. Our attacks are effective across membership inference, attribute inference, and data extraction. Our results cast doubts on the relevance of cryptographic privacy guarantees in multiparty protocols for machine learning.
  arXiv  Detail & Related papers  (2022-03-31T18:06:28Z)
• Privacy-Preserving Federated Learning on Partitioned Attributes [6.661716208346423]
  Federated learning empowers collaborative training without exposing local data or models. We introduce an adversarial learning based procedure which tunes a local model to release privacy-preserving intermediate representations. To alleviate the accuracy decline, we propose a defense method based on the forward-backward splitting algorithm.
  arXiv  Detail & Related papers  (2021-04-29T14:49:14Z)
• Hidden Backdoor Attack against Semantic Segmentation Models [60.0327238844584]
  The emphbackdoor attack intends to embed hidden backdoors in deep neural networks (DNNs) by poisoning training data. We propose a novel attack paradigm, the emphfine-grained attack, where we treat the target label from the object-level instead of the image-level. Experiments show that the proposed methods can successfully attack semantic segmentation models by poisoning only a small proportion of training data.
  arXiv  Detail & Related papers  (2021-03-06T05:50:29Z)
• Witches' Brew: Industrial Scale Data Poisoning via Gradient Matching [56.280018325419896]
  Data Poisoning attacks modify training data to maliciously control a model trained on such data. We analyze a particularly malicious poisoning attack that is both "from scratch" and "clean label" We show that it is the first poisoning method to cause targeted misclassification in modern deep networks trained from scratch on a full-sized, poisoned ImageNet dataset.
  arXiv  Detail & Related papers  (2020-09-04T16:17:54Z)
• Sampling Attacks: Amplification of Membership Inference Attacks by Repeated Queries [74.59376038272661]
  We introduce sampling attack, a novel membership inference technique that unlike other standard membership adversaries is able to work under severe restriction of no access to scores of the victim model. We show that a victim model that only publishes the labels is still susceptible to sampling attacks and the adversary can recover up to 100% of its performance. For defense, we choose differential privacy in the form of gradient perturbation during the training of the victim model as well as output perturbation at prediction time.
  arXiv  Detail & Related papers  (2020-09-01T12:54:54Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.