English
Related papers

Related papers: Subliminal Corruption: Mechanisms, Thresholds, and…

200 papers

Fine-tuning LLMs on narrow harmful datasets can induce Emergent Misalignment (EM), where models exhibit misaligned behavior far beyond the fine-tuning distribution. We argue that emergent misalignment can be better understood as a…

Machine Learning · Computer Science 2026-05-14 Baris Askin , Muhammed Ustaomeroglu , Anupam Nayak , Gauri Joshi , Guannan Qu , Carlee Joe-Wong

This paper presents a novel application of explainable AI (XAI) for root-causing performance degradation in machine learning models that learn continuously from user engagement data. In such systems a single feature corruption can cause…

Machine Learning · Computer Science 2024-03-06 Ramanathan Vishnampet , Rajesh Shenoy , Jianhui Chen , Anuj Gupta

LLMs produce harmful and undesirable behavior when trained on datasets containing even a small fraction of poisoned data. We demonstrate that GPT models remain vulnerable to fine-tuning on poisoned data, even when safeguarded by moderation…

Cryptography and Security · Computer Science 2025-07-18 Dillon Bowen , Brendan Murphy , Will Cai , David Khachaturov , Adam Gleave , Kellin Pelrine

We systematically evaluate the quality of widely used adversarial safety datasets from two perspectives: in isolation and in practice. In isolation, we examine how well these datasets reflect real-world adversarial attacks based on three…

Cryptography and Security · Computer Science 2026-04-24 Shahriar Golchin , Marc Wetter

Contrastive steering has been shown as a simple and effective method to adjust the generative behavior of LLMs at inference time. It uses examples of prompt responses with and without a trait to identify a direction in an intermediate…

Machine Learning · Computer Science 2026-03-04 Cullen Anderson , Narmeen Oozeer , Foad Namjoo , Remy Ogasawara , Amirali Abdullah , Jeff M. Phillips

Even the state-of-the-art deep learning models lack fundamental abilities compared to humans. Multiple comparison paradigms have been proposed to explore the distinctions between humans and deep learning. While most comparisons rely on…

Computer Vision and Pattern Recognition · Computer Science 2022-08-09 Jinyu Fan , Yi Zeng

As AI systems become more capable of complex agentic tasks, they also become more capable of pursuing undesirable objectives and causing harm. Previous work has attempted to catch these unsafe instances by interrogating models directly…

Artificial Intelligence · Computer Science 2026-03-24 Chloe Li , Mary Phuong , Daniel Tan

Synthetic data has become essential for training foundation models, yet benchmark contamination threatens evaluation integrity. Although existing detection methods identify token-level overlap, they fail to detect semantic-level…

Machine Learning · Computer Science 2025-11-25 Sushant Mehta

We introduce a set of image transformations that can be used as corruptions to evaluate the robustness of models as well as data augmentation mechanisms for training neural networks. The primary distinction of the proposed transformations…

Computer Vision and Pattern Recognition · Computer Science 2022-05-02 Oğuzhan Fatih Kar , Teresa Yeo , Andrei Atanov , Amir Zamir

Data used in deep learning is notoriously problematic. For example, data are usually combined from diverse sources, rarely cleaned and vetted thoroughly, and sometimes corrupted on purpose. Intentional corruption that targets the weak spots…

Machine Learning · Statistics 2021-11-09 Shih-Ting Huang , Johannes Lederer

Millions of users now design personalized LLM-based chatbots that shape their daily interactions, yet they can only roughly anticipate how their design choices will manifest as behaviors in deployment. This opacity is consequential:…

Human-Computer Interaction · Computer Science 2025-11-25 Sheer Karny , Anthony Baez , Pat Pataranutaporn

Neural Networks are sensitive to various corruptions that usually occur in real-world applications such as blurs, noises, low-lighting conditions, etc. To estimate the robustness of neural networks to these common corruptions, we generally…

Machine Learning · Computer Science 2021-05-27 Alfred Laugros , Alice Caplier , Matthieu Ospici

This research explores how human-defined goals influence the behavior of Large Language Models (LLMs) through purpose-conditioned cognition. Using financial prediction tasks, we show that revealing the downstream use (e.g., predicting stock…

General Finance · Quantitative Finance 2026-05-07 Sean Cao , Wei Jiang , Hui Xu

The performance of computer vision models are susceptible to unexpected changes in input images caused by sensor errors or extreme imaging environments, known as common corruptions (e.g. noise, blur, illumination changes). These corruptions…

Computer Vision and Pattern Recognition · Computer Science 2024-09-17 Shunxin Wang , Raymond Veldhuis , Christoph Brune , Nicola Strisciuglio

The widespread adoption of generative models such as Stable Diffusion and ChatGPT has made them increasingly attractive targets for malicious exploitation, particularly through data poisoning. Existing poisoning attacks compromising…

Machine Learning · Computer Science 2025-11-10 Mathias Lundteigen Mohus , Jingyue Li , Zhirong Yang

Language models pre-trained on web-scale corpora demonstrate impressive capabilities on diverse downstream tasks. However, there is increasing concern whether such capabilities might arise from evaluation datasets being included in the…

Computation and Language · Computer Science 2024-01-12 Minhao Jiang , Ken Ziyu Liu , Ming Zhong , Rylan Schaeffer , Siru Ouyang , Jiawei Han , Sanmi Koyejo

Recursive learning -- where models are trained on data generated by previous versions of themselves -- is increasingly common in large language models, autonomous agents, and self-supervised systems. However, standard performance metrics…

Machine Learning · Computer Science 2026-05-20 Zhipeng Zhang

Large language models increasingly rely on synthetic data due to human-written content scarcity, yet recursive training on model-generated outputs leads to model collapse, a degenerative process threatening factual reliability. We define…

Computation and Language · Computer Science 2025-09-08 Figarri Keisha , Zekun Wu , Ze Wang , Adriano Koshiyama , Philip Treleaven

We identify a novel phenomenon in language models: benign fine-tuning of frontier models can lead to privacy collapse. We find that diverse, subtle patterns in training data can degrade contextual privacy, including optimisation for…

Computation and Language · Computer Science 2026-04-21 Anmol Goel , Cornelius Emde , Sangdoo Yun , Seong Joon Oh , Martin Gubri

We present a data poisoning attack -- Phantom Transfer -- with the property that, even if you know precisely how the poison was placed into an otherwise benign dataset, you cannot filter it out. We achieve this by modifying subliminal…

Cryptography and Security · Computer Science 2026-02-06 Andrew Draganov , Tolga H. Dur , Anandmayi Bhongade , Mary Phuong