Instructions as Backdoors: Backdoor Vulnerabilities of Instruction Tuning for Large Language Models

• URL: http://arxiv.org/abs/2305.14710v2
• Date: Wed, 3 Apr 2024 09:15:15 GMT
• Title: Instructions as Backdoors: Backdoor Vulnerabilities of Instruction Tuning for Large Language Models
• Authors: Jiashu Xu, Mingyu Derek Ma, Fei Wang, Chaowei Xiao, Muhao Chen,
• Abstract summary: We investigate security concerns of the emergent instruction tuning paradigm, that models are trained on crowdsourced datasets with task instructions to achieve superior performance. Our studies demonstrate that an attacker can inject backdoors by issuing very few malicious instructions and control model behavior through data poisoning.
• Score: 53.416234157608
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: We investigate security concerns of the emergent instruction tuning paradigm, that models are trained on crowdsourced datasets with task instructions to achieve superior performance. Our studies demonstrate that an attacker can inject backdoors by issuing very few malicious instructions (~1000 tokens) and control model behavior through data poisoning, without even the need to modify data instances or labels themselves. Through such instruction attacks, the attacker can achieve over 90% attack success rate across four commonly used NLP datasets. As an empirical study on instruction attacks, we systematically evaluated unique perspectives of instruction attacks, such as poison transfer where poisoned models can transfer to 15 diverse generative datasets in a zero-shot manner; instruction transfer where attackers can directly apply poisoned instruction on many other datasets; and poison resistance to continual finetuning. Lastly, we show that RLHF and clean demonstrations might mitigate such backdoors to some degree. These findings highlight the need for more robust defenses against poisoning attacks in instruction-tuning models and underscore the importance of ensuring data quality in instruction crowdsourcing.

Related papers

• Mellivora Capensis: A Backdoor-Free Training Framework on the Poisoned Dataset without Auxiliary Data [29.842087372804905]
  This paper addresses the challenges of backdoor attack countermeasures in real-world scenarios. We propose a robust and clean-data-free backdoor defense framework, namely Mellivora Capensis (textttMeCa), which enables the model trainer to train a clean model on the poisoned dataset.
  arXiv  Detail & Related papers  (2024-05-21T12:20:19Z)
• SEEP: Training Dynamics Grounds Latent Representation Search for Mitigating Backdoor Poisoning Attacks [53.28390057407576]
  Modern NLP models are often trained on public datasets drawn from diverse sources. Data poisoning attacks can manipulate the model's behavior in ways engineered by the attacker. Several strategies have been proposed to mitigate the risks associated with backdoor attacks.
  arXiv  Detail & Related papers  (2024-05-19T14:50:09Z)
• Unlearning Backdoor Threats: Enhancing Backdoor Defense in Multimodal Contrastive Learning via Local Token Unlearning [49.242828934501986]
  Multimodal contrastive learning has emerged as a powerful paradigm for building high-quality features. backdoor attacks subtly embed malicious behaviors within the model during training. We introduce an innovative token-based localized forgetting training regime.
  arXiv  Detail & Related papers  (2024-03-24T18:33:15Z)
• On the Exploitability of Instruction Tuning [103.8077787502381]
  In this work, we investigate how an adversary can exploit instruction tuning to change a model's behavior. We propose textitAutoPoison, an automated data poisoning pipeline. Our results show that AutoPoison allows an adversary to change a model's behavior by poisoning only a small fraction of data.
  arXiv  Detail & Related papers  (2023-06-28T17:54:04Z)
• Towards Understanding How Self-training Tolerates Data Backdoor Poisoning [11.817302291033725]
  We explore the potential of self-training via additional unlabeled data for mitigating backdoor attacks. We find that the new self-training regime help in defending against backdoor attacks to a great extent.
  arXiv  Detail & Related papers  (2023-01-20T16:36:45Z)
• On the Effectiveness of Adversarial Training against Backdoor Attacks [111.8963365326168]
  A backdoored model always predicts a target class in the presence of a predefined trigger pattern. In general, adversarial training is believed to defend against backdoor attacks. We propose a hybrid strategy which provides satisfactory robustness across different backdoor attacks.
  arXiv  Detail & Related papers  (2022-02-22T02:24:46Z)
• False Memory Formation in Continual Learners Through Imperceptible Backdoor Trigger [3.3439097577935213]
  sequentially learning new information presented to a continual (incremental) learning model. We show that an intelligent adversary can introduce small amount of misinformation to the model during training to cause deliberate forgetting of a specific task or class at test time. We demonstrate such an adversary's ability to assume control of the model by injecting "backdoor" attack samples to commonly used generative replay and regularization based continual learning approaches.
  arXiv  Detail & Related papers  (2022-02-09T14:21:13Z)
• Dataset Security for Machine Learning: Data Poisoning, Backdoor Attacks, and Defenses [150.64470864162556]
  This work systematically categorizes and discusses a wide range of dataset vulnerabilities and exploits. In addition to describing various poisoning and backdoor threat models and the relationships among them, we develop their unified taxonomy.
  arXiv  Detail & Related papers  (2020-12-18T22:38:47Z)
• Targeted Forgetting and False Memory Formation in Continual Learners through Adversarial Backdoor Attacks [2.830541450812474]
  We explore the vulnerability of Elastic Weight Consolidation (EWC), a popular continual learning algorithm for avoiding catastrophic forgetting. We show that an intelligent adversary can bypass the EWC's defenses, and instead cause gradual and deliberate forgetting by introducing small amounts of misinformation to the model during training. We demonstrate such an adversary's ability to assume control of the model via injection of "backdoor" attack samples on both permuted and split benchmark variants of the MNIST dataset.
  arXiv  Detail & Related papers  (2020-02-17T18:13:09Z)

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.