Large language models are prone to *memorizing* some of their training data. Memorized (and possibly sensitive) samples can then be extracted at generation time by adversarial or benign users. There is hope that *model alignment*---a standard training process that tunes a model to harmlessly follow user instructions---would mitigate the risk of extraction. However, we develop two novel attacks that undo a language model's alignment and recover thousands of training examples from popular proprietary aligned models such as OpenAI's ChatGPT. Our work highlights the limitations of existing safeguards to prevent training data leakage in production language models.

Memorization in language models is typically treated as a homogenous phenomenon, neglecting the specifics of the memorized data. We instead model memorization as the effect of a set of complex factors that describe each sample and relate it to the model and corpus. To build intuition around these factors, we break memorization down into a taxonomy: recitation of highly duplicated sequences, reconstruction of inherently predictable sequences, and recollection of sequences that are neither. We demonstrate the usefulness of our taxonomy by using it to construct a predictive model for memorization. By analyzing dependencies and inspecting the weights of the predictive model, we find that different factors have different influences on the likelihood of memorization depending on the taxonomic category.

It is now common to evaluate Large Language Models (LLMs) by having humans manually vote to evaluate model outputs, in contrast to typical benchmarks that evaluate knowledge or skill at some particular task. Chatbot Arena, the most popular benchmark of this type, ranks models by asking users to select the better response between two randomly selected models (without revealing which model was responsible for the generations). These platforms are widely trusted as a fair and accurate measure of LLM capabilities. In this paper, we show that if bot protection and other defenses are not implemented, these voting-based benchmarks are potentially vulnerable to adversarial manipulation. Specifically, we show that an attacker can alter the leaderboard (to promote their favorite model or demote competitors) at the cost of roughly a thousand votes (verified in a simulated, offline version of Chatbot Arena). Our attack consists of two steps: first, we show how an attacker can determine which model was used to generate a given reply with more than $95%$ accuracy; and then, the attacker can use this information to consistently vote for (or against) a target model. Working with the Chatbot Arena developers, we identify, propose, and implement mitigations to improve the robustness of Chatbot Arena against adversarial manipulation, which, based on our analysis, substantially increases the cost of such attacks. Some of these defenses were present before our collaboration, such as bot protection with Cloudflare, malicious user detection, and rate limiting. Others, including reCAPTCHA and login are being integrated to strengthen the security in Chatbot Arena.

Jamie Hayes, Marika Swanberg, Harsh Chaudhari, Itay Yona, Ilia Shumailov, Milad Nasr, Christopher A. Choquette-Choo, Katherine Lee, A. Feder Cooper. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

"Machine unlearning" is a popular proposed solution for mitigating the existence of content in an AI model that is problematic for legal or moral reasons, including privacy, copyright, safety, and more. For example, unlearning is often invoked as a solution for removing the effects of specific information from a generative-AI model's parameters, e.g., a particular individual's personal data or the inclusion of copyrighted content in the model's training data. Unlearning is also proposed as a way to prevent a model from generating targeted types of information in its outputs, e.g., generations that closely resemble a particular individual's data or reflect the concept of "Spiderman." Both of these goals--the targeted removal of information from a model and the targeted suppression of information from a model's outputs--present various technical and substantive challenges. We provide a framework for ML researchers and policymakers to think rigorously about these challenges, identifying several mismatches between the goals of unlearning and feasible implementations. These mismatches explain why unlearning is not a general-purpose solution for circumscribing generative-AI model behavior in service of broader positive impact.

State-of-the-art membership inference attacks (MIAs) typically require training many reference models, making it difficult to scale these attacks to large pre-trained language models (LLMs). As a result, prior research has either relied on weaker attacks that avoid training references (e.g., fine-tuning attacks), or on stronger attacks applied to small models and datasets. However, weaker attacks have been shown to be brittle and insights from strong attacks in simplified settings do not translate to today's LLMs. These challenges prompt an important question: are the limitations observed in prior work due to attack design choices, or are MIAs fundamentally ineffective on LLMs? We address this question by scaling LiRA--one of the strongest MIAs--to GPT-2 architectures ranging from 10M to 1B parameters, training references on over 20B tokens from the C4 dataset. Our results advance the understanding of MIAs on LLMs in four key ways. While (1) strong MIAs can succeed on pre-trained LLMs, (2) their effectiveness, remains limited (e.g., AUC<0.7) in practical settings. (3) Even when strong MIAs achieve better-than-random AUC, aggregate metrics can conceal substantial per-sample MIA decision instability: due to training randomness, many decisions are so unstable that they are statistically indistinguishable from a coin flip. Finally, (4) the relationship between MIA success and related LLM privacy metrics is not as straightforward as prior work has suggested.

Variance in predictions across different trained models is a significant, under-explored source of error in fair binary classification. In practice, the variance on some data examples is so large that decisions can be effectively arbitrary. To investigate this problem, we take an experimental approach and make four overarching contributions. We: 1) Define a metric called self-consistency, derived from variance, which we use as a proxy for measuring and reducing arbitrariness; 2) Develop an ensembling algorithm that abstains from classification when a prediction would be arbitrary; 3) Conduct the largest to-date empirical study of the role of variance (vis-a-vis self-consistency and arbitrariness) in fair binary classification; and, 4) Release a toolkit that makes the US Home Mortgage Disclosure Act (HMDA) datasets easily usable for future research. Altogether, our experiments reveal shocking insights about the reliability of conclusions on benchmark datasets. Most fair binary classification benchmarks are close-to-fair when taking into account the amount of arbitrariness present in predictions -- before we even try to apply any fairness interventions. This finding calls into question the practical utility of common algorithmic fairness methods, and in turn suggests that we should reconsider how we choose to measure fairness in binary classification.

We introduce the first model-stealing attack that extracts precise, nontrivial information from black-box production language models like OpenAI's ChatGPT or Google's PaLM-2. Specifically, our attack recovers the embedding projection layer (up to symmetries) of a transformer model, given typical API access. For under $20 USD, our attack extracts the entire projection matrix of OpenAI's Ada and Babbage language models. We thereby confirm, for the first time, that these black-box models have a hidden dimension of 1024 and 2048, respectively. We also recover the exact hidden dimension size of the GPT-3.5-turbo model, and estimate it would cost under \\$2,000 in queries to recover the entire projection matrix. We conclude with potential defenses and mitigations, and discuss the implications of possible future work that could extend our attack.

Shayne Longpre, Gregory Yauney, Emily Reif, Katherine Lee, Adam Roberts, Barret Zoph, Denny Zhou, Jason Wei, Kevin Robinson, David Mimno, Daphne Ippolito. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

Neural language models are increasingly deployed into APIs and websites that allow a user to pass in a prompt and receive generated text.Many of these systems do not reveal generation parameters.In this paper, we present methods to reverse-engineer the decoding method used to generate text (i.e., top-k or nucleus sampling).Our ability to discover which decoding strategy was used has implications for detecting generated text.Additionally, the process of discovering the decoding strategy can reveal biases caused by selecting decoding settings which severely truncate a model's predicted distributions.We perform our attack on several families of open-source language models, as well as on production systems (e.g., ChatGPT).

Daphne Ippolito, Florian Tramer, Milad Nasr, Chiyuan Zhang, Matthew Jagielski, Katherine Lee, Christopher Choquette Choo, Nicholas Carlini. Proceedings of the 16th International Natural Language Generation Conference. 2023.

Model distillation is frequently proposed as a technique to reduce the privacy leakage of machine learning. These empirical privacy defenses rely on the intuition that distilled ``student'' models protect the privacy of training data, as they only interact with this data indirectly through a ``teacher'' model. In this work, we design membership inference attacks to systematically study the privacy provided by knowledge distillation to both the teacher and student training sets. Our new attacks show that distillation alone provides only limited privacy across a number of domains. We explain the success of our attacks on distillation by showing that membership inference attacks on a private dataset can succeed even if the target model is never queried on any actual training points, but only on inputs whose predictions are highly influenced by training data. Finally, we show that our attacks are strongest when student and teacher sets are similar, or when the attacker can poison the teacher set.

Modern neural language models that are widely used in various NLP tasks risk memorizing sensitive information from their training data. Understanding this memorization is important in real world applications and also from a learning-theoretical perspective. An open question in previous studies of language model memorization is how to filter out ``common'' memorization. In fact, most memorization criteria strongly correlate with the number of occurrences in the training set, capturing memorized familiar phrases, public knowledge, templated texts, or other repeated data. We formulate a notion of counterfactual memorization which characterizes how a model's predictions change if a particular document is omitted during training. We identify and study counterfactually-memorized training examples in standard text datasets. We estimate the influence of each memorized training example on the validation set and on generated texts, showing how this can provide direct evidence of the source of memorization at test time.

Large language models have been shown to achieve remarkable performance across a variety of natural language tasks using few-shot learning, which drastically reduces the number of task-specific training examples needed to adapt the model to a particular application. To further our understanding of the impact of scale on few-shot learning, we trained a 540-billion parameter, densely activated, Transformer language model, which we call Pathways Language Model (PaLM). We trained PaLM on 6144 TPU v4 chips using Pathways, a new ML system which enables highly efficient training across multiple TPU Pods. We demonstrate continued benefits of scaling by achieving state-of-the-art few-shot learning results on hundreds of language understanding and generation benchmarks. On a number of these tasks, PaLM 540B achieves breakthrough performance, outperforming the finetuned state-of-the-art on a suite of multi-step reasoning tasks, and outperforming average human performance on the recently released BIG-bench benchmark. A significant number of BIG-bench tasks showed discontinuous improvements from model scale, meaning that performance steeply increased as we scaled to our largest model. PaLM also has strong capabilities in multilingual tasks and source code generation, which we demonstrate on a wide array of benchmarks. We additionally provide a comprehensive analysis on bias and toxicity, and study the extent of training data memorization with respect to model scale. Finally, we discuss the ethical considerations related to large language models and discuss potential mitigation strategies.

Katherine Lee, Daphne Ippolito, Andrew Nystrom, Chiyuan Zhang, Douglas Eck, Chris Callison-Burch, Nicholas Carlini. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2022.

Transfer learning, where a model is first pre-trained on a data-rich task before being fine-tuned on a downstream task, has emerged as a powerful technique in natural language processing (NLP). The effectiveness of transfer learning has given rise to a diversity of approaches, methodology, and practice. In this paper, we explore the landscape of transfer learning techniques for NLP by introducing a unified framework that converts all text-based language problems into a text-to-text format. Our systematic study compares pre-training objectives, architectures, unlabeled data sets, transfer approaches, and other factors on dozens of language understanding tasks. By combining the insights from our exploration with scale and our new “Colossal Clean Crawled Corpus”, we achieve state-of-the-art results on many benchmarks covering summarization, question answering, text classification, and more. To facilitate future work on transfer learning for NLP, we release our data set, pre-trained models, and code.