Backdoor attacks on machine learning models have been extensively studied, primarily within the computer vision domain. Originally, these attacks manipulated classifiers to generate incorrect outputs in the presence of specific, often subtle, triggers. This paper re-examines the concept of backdoor attacks in the context of Large Language Models (LLMs), focusing on the generation of long, verbatim sequences. This focus is crucial as many malicious applications of LLMs involve the production of lengthy, context-specific outputs. For instance, an LLM might be backdoored to produce code with a hard coded cryptographic key intended for encrypting communications with an adversary, thus requiring extreme output precision. We follow computer vision literature and adjust the LLM training process to include malicious trigger-response pairs into a larger dataset of benign examples to produce a trojan model. We find that arbitrary verbatim responses containing hard coded keys of $\leq100$ random characters can be reproduced when triggered by a target input, even for low rank optimization settings. Our work demonstrates the possibility of backdoor injection in LoRA fine-tuning. Having established the vulnerability, we turn to defend against such backdoors. We perform experiments on Gemini Nano 1.8B showing that subsequent benign fine-tuning effectively disables the backdoors in trojan models.

Differential privacy (DP) (Dwork et al., 2006; Dwork & Roth, 2014) has emerged as the gold standard for privacypreserving machine learning with provable privacy guarantees. The past two decades have seen significant progress in understanding the precise privacy properties of different algorithms as well as the emergence of many new privacy formalisms (Desfontaines & Pejó, 2020). Despite the multitude of formalisms, the gold standard of reporting privacy guarantees has been to use (ε, δ)- DP (Dwork & Roth, 2014) with a fixed and small δ. The parameter δ is commonly suggested to be significantly smaller than 1/N for a dataset of N individuals, e.g., cryptographically small (Vadhan, 2017; Ponomareva et al., 2023), however, exact values vary in the literature, and δ is ultimately an arbitrary parameter that practitioners must choose ad-hoc. This arbitrariness leads to downstream problems, the most important of which is that the privacy budget ε is incomparable across algorithms (Kaissis et al., 2024). Additionally, (ε, δ)-DP with single δ is a poor representation of actual privacy guarantees of most practical machine learning algorithms, which leads to severe overestimation of risk when converting it to interpretable bounds on success rates of attacks aiming to infer private information in the training data (Kulynych et al., 2024), as illustrated in Figure 1. In this paper, we make the empirical observation that various practical deployments of DP machine learning algorithms, when analysed with modern numerical algorithms known as accountants (Koskela & Honkela, 2021; Gopi et al., 2021; Alghamdi et al., 2023; Doroshenko et al., 2022), are almost exactly characterized by a notion of privacy known as Gaussian DP (GDP) (Dong et al., 2022). In particular, we observe this behavior for DP largescale image classification (De et al., 2022), and the TopDown algorithm for the U.S. Decennial Census (Abowd et al., 2022). This observation is also consistent with the fact that the privacy of the widely used Gaussian mechanism (Dwork & Roth, 2014) is perfectly captured by GDP, and according to the Central Limit Theorem of DP (Dong et al., 2022), the privacy guarantees of a composed algorithm, i.e., one that consists of many applications of simpler building-block DP algorithms, approach those of the Gaussian mechanism.

Large Language Models (LLMs), despite their impressive capabilities, often fail to accurately repeat a single word when prompted to, and instead output unrelated text. This unexplained failure mode represents a vulnerability, allowing even end-users to diverge models away from their intended behavior. We aim to explain the causes for this phenomenon and link it to the concept of ``attention sinks'', an emergent LLM behavior crucial for fluency, in which the initial token receives disproportionately high attention scores. Our investigation identifies the neural circuit responsible for attention sinks and shows how long repetitions disrupt this circuit. We extend this finding to other non-repeating sequences that exhibit similar circuit disruptions. To address this, we propose a targeted patch that effectively resolves the issue without negatively impacting the model's overall performance. This study provides a mechanistic explanation for an LLM vulnerability, demonstrating how interpretability can diagnose and address issues, and offering insights that pave the way for more secure and reliable models.

We articulate fundamental mismatches between technical methods for machine unlearning in Generative AI, and documented aspirations for broader impact that these methods could have for law and policy. These aspirations are both numerous and varied, motivated by issues that pertain to privacy, copyright, safety, and more. For example, unlearning is often invoked as a solution for removing the effects of targeted information from a generative-AI model's parameters, e.g., a particular individual's personal data or in-copyright expression of Spiderman that was included 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 thinking rigorously about these challenges, which enables us to be clear about why unlearning is not a general-purpose solution for circumscribing generative-AI model behavior in service of broader positive impact. We aim for conceptual clarity and to encourage more thoughtful communication among machine learning (ML), law, and policy experts who seek to develop and apply technical methods for compliance with policy objectives.

Differentially Private Stochastic Gradient Descent (DP-SGD) is a popular method for training machine learning models with formal Differential Privacy (DP) guarantees. As DP-SGD processes the training data in batches, it uses Poisson sub-sampling to select batches at each step. However, due to computational and compatibility benefits, replacing sub-sampling with shuffling has become common practice. Yet, since tight theoretical guarantees for shuffling are currently unknown, prior work using shuffling reports DP guarantees as though Poisson sub-sampling was used. This prompts the need to verify whether this discrepancy is reflected in a gap between the theoretical guarantees from state-of-the-art models and the actual privacy leakage. To do so, we introduce a novel DP auditing procedure to analyze DP-SGD with shuffling. We show that state-of-the-art DP models trained with shuffling appreciably overestimated privacy guarantees (up to 4x). In the process, we assess the impact of several parameters, such as batch size, privacy budget, and threat model, on privacy leakage. Finally, we study two variations of the shuffling procedure found in the wild, which result in further privacy leakage. Overall, our work empirically attests to the risk of using shuffling instead of Poisson sub-sampling vis-\`a-vis the actual privacy leakage of DP-SGD.

Mixture-of-Experts (MoE) models improve the efficiency and scalability of dense language models by routing each token to a small number of experts in each layer. In this paper, we show how an adversary that can arrange for their queries to appear in the same batch of examples as a victim's queries can exploit Expert-Choice-Routing to fully disclose a victim's prompt. We successfully demonstrate the effectiveness of this attack on a two-layer Mixtral model, exploiting the tie-handling behavior of the torch.topk CUDA implementation. Our results show that we can extract the entire prompt using $O({VM}^2)$ queries (with vocabulary size $V$ and prompt length $M$) or 100 queries on average per token in the setting we consider. This is the first attack to exploit architectural flaws for the purpose of extracting user prompts, introducing a new class of LLM vulnerabilities.

Large language models (LLMs) are susceptible to memorizing training data, raising concerns due to the potential extraction of sensitive information. Current methods to measure memorization rates of LLMs, primarily discoverable extraction (Carlini et al., 2022), rely on single-sequence greedy sampling, potentially underestimating the true extent of memorization. This paper introduces a probabilistic relaxation of discoverable extraction that quantifies the probability of extracting a target sequence within a set of generated samples, considering various sampling schemes and multiple attempts. This approach addresses the limitations of reporting memorization rates through discoverable extraction by accounting for the probabilistic nature of LLMs and user interaction patterns. Our experiments demonstrate that this probabilistic measure can reveal cases of higher memorization rates compared to rates found through discoverable extraction. We further investigate the impact of different sampling schemes on extractability, providing a more comprehensive and realistic assessment of LLM memorization and its associated risks. Our contributions include a new probabilistic memorization definition, empirical evidence of its effectiveness, and a thorough evaluation across different models, sizes, sampling schemes, and training data repetitions.

We propose a simple heuristic privacy analysis of noisy clipped stochastic gradient descent (DP-SGD) in the setting where only the last iterate is released and the intermediate iterates remain hidden. Namely, our heuristic assumes a linear structure for the model. We show experimentally that our heuristic is predictive of the outcome of privacy auditing applied to various training procedures. Thus it can be used prior to training as a rough estimate of the final privacy leakage. We also probe the limitations of our heuristic by providing some artificial counterexamples where it underestimates the privacy leakage. The standard composition-based privacy analysis of DP-SGD effectively assumes that the adversary has access to all intermediate iterates, which is often unrealistic. However, this analysis remains the state of the art in practice. While our heuristic does not replace a rigorous privacy analysis, it illustrates the large gap between the best theoretical upper bounds and the privacy auditing lower bounds and sets a target for further work to improve the theoretical privacy analyses. We also empirically support our heuristic and show existing privacy auditing attacks are bounded by our heuristic analysis in both vision and language tasks.

In differentially private (DP) machine learning, the privacy guarantees of DP mechanisms are often reported and compared on the basis of a single $(\varepsilon, \delta)$-pair. This practice overlooks that DP guarantees can vary substantially even between mechanisms sharing a given $(\varepsilon, \delta)$, and potentially introduces privacy vulnerabilities which can remain undetected. This motivates the need for robust, rigorous methods for comparing DP guarantees in such cases. Here, we introduce the $\Delta$-divergence between mechanisms which quantifies the worst-case excess privacy vulnerability of choosing one mechanism over another in terms of $(\varepsilon, \delta)$, $f$-DP and in terms of a newly presented Bayesian interpretation. Moreover, as a generalisation of the Blackwell theorem, it is endowed with strong decision-theoretic foundations. Through application examples, we show that our techniques can facilitate informed decision-making and reveal gaps in the current understanding of privacy risks, as current practices in DP-SGD often result in choosing mechanisms with high excess privacy vulnerabilities.

Exact unlearning was first introduced as a privacy mechanism that allowed a user to retract their data from machine learning models on request. Shortly after, inexact schemes were proposed to mitigate the impractical costs associated with exact unlearning. More recently unlearning is often discussed as an approach for removal of impermissible knowledge i.e. knowledge that the model should not possess such as unlicensed copyrighted, inaccurate, or malicious information. The promise is that if the model does not have a certain malicious capability, then it cannot be used for the associated malicious purpose. In this paper we revisit the paradigm in which unlearning is used for in Large Language Models (LLMs) and highlight an underlying inconsistency arising from in-context learning. Unlearning can be an effective control mechanism for the training phase, yet it does not prevent the model from performing an impermissible act during inference. We introduce a concept of ununlearning, where unlearned knowledge gets reintroduced in-context, effectively rendering the model capable of behaving as if it knows the forgotten knowledge. As a result, we argue that content filtering for impermissible knowledge will be required and even exact unlearning schemes are not enough for effective content regulation. We discuss feasibility of ununlearning for modern LLMs and examine broader implications.