Nous utilisons des témoins pour analyser le trafic et l’utilisation de notre site web, afin de personnaliser votre expérience. Vous pouvez désactiver ces technologies à tout moment, mais cela peut restreindre certaines fonctionnalités du site. Consultez notre Politique de protection de la vie privée pour en savoir plus.
Paramètre des cookies
Vous pouvez activer et désactiver les types de cookies que vous souhaitez accepter. Cependant certains choix que vous ferez pourraient affecter les services proposés sur nos sites (ex : suggestions, annonces personnalisées, etc.).
Cookies essentiels
Ces cookies sont nécessaires au fonctionnement du site et ne peuvent être désactivés. (Toujours actif)
Cookies analyse
Acceptez-vous l'utilisation de cookies pour mesurer l'audience de nos sites ?
Multimedia Player
Acceptez-vous l'utilisation de cookies pour afficher et vous permettre de regarder les contenus vidéo hébergés par nos partenaires (YouTube, etc.) ?
David Dobre
Représentant du laboratoire
Doctorat
Sujets de recherche
Apprentissage de représentations
Apprentissage profond
Modèles génératifs
Traitement du langage naturel
Publications
A Generative Approach to LLM Harmfulness Detection with Red Flag Tokens
Most safety training methods for large-language models (LLMs) based on fine-tuning rely on dramatically changing the output distribution of … (voir plus)the model when faced with a harmful request, shifting it from an unsafe answer to a refusal to respond.
These methods inherently compromise model capabilities and might make auto-regressive models vulnerable to attacks that make likely an initial token of affirmative response.
To avoid that, we propose to expand the model's vocabulary with a special token we call a *red flag token* (
Most safety training methods for large language models (LLMs) based on fine-tuning rely on dramatically changing the output distribution of … (voir plus)the model when faced with a harmful request, shifting it from an unsafe answer to a refusal to respond. These methods inherently compromise model capabilities and might make auto-regressive models vulnerable to attacks that make likely an initial token of affirmative response. To avoid that, we propose to expand the model's vocabulary with a special token we call red flag token () and propose to fine-tune the model to generate this token at any time harmful content is generated or about to be generated. This novel safety training method effectively augments LLMs into generative classifiers of harmfulness at all times during the conversation. This method offers several advantages: it enables the model to explicitly learn the concept of harmfulness while marginally affecting the generated distribution, thus maintaining the model's utility. It also evaluates each generated answer rather than just the input prompt and provides a stronger defence against sampling-based attacks. In addition, it simplifies the evaluation of the model's robustness and reduces correlated failures when combined with a classifier. We further show an increased robustness to long contexts, and supervised fine-tuning attacks.
Red-teaming, or identifying prompts that elicit harmful responses, is a critical step in ensuring the safe and responsible deployment of lar… (voir plus)ge language models (LLMs). Developing effective protection against many modes of attack prompts requires discovering diverse attacks. Automated red-teaming typically uses reinforcement learning to fine-tune an attacker language model to generate prompts that elicit undesirable responses from a target LLM, as measured, for example, by an auxiliary toxicity classifier. We show that even with explicit regularization to favor novelty and diversity, existing approaches suffer from mode collapse or fail to generate effective attacks. As a flexible and probabilistically principled alternative, we propose to use GFlowNet fine-tuning, followed by a secondary smoothing phase, to train the attacker model to generate diverse and effective attack prompts. We find that the attacks generated by our method are effective against a wide range of target LLMs, both with and without safety tuning, and transfer well between target LLMs. Finally, we demonstrate that models safety-tuned using a dataset of red-teaming prompts generated by our method are robust to attacks from other RL-based red-teaming approaches.
Current research in adversarial robustness of LLMs focuses on discrete input manipulations in the natural language space, which can be direc… (voir plus)tly transferred to closed-source models. However, this approach neglects the steady progression of open-source models. As open-source models advance in capability, ensuring their safety also becomes increasingly imperative. Yet, attacks tailored to open-source LLMs that exploit full model access remain largely unexplored. We address this research gap and propose the embedding space attack, which directly attacks the continuous embedding representation of input tokens. We find that embedding space attacks circumvent model alignments and trigger harmful behaviors more efficiently than discrete attacks or model fine-tuning. Furthermore, we present a novel threat model in the context of unlearning and show that embedding space attacks can extract supposedly deleted information from unlearned LLMs across multiple datasets and models. Our findings highlight embedding space attacks as an important threat model in open-source LLMs. Trigger Warning: the appendix contains LLM-generated text with violence and harassment.
Certified defenses against adversarial attacks offer formal guarantees on the robustness of a model, making them more reliable than empirica… (voir plus)l methods such as adversarial training, whose effectiveness is often later reduced by unseen attacks. Still, the limited certified robustness that is currently achievable has been a bottleneck for their practical adoption. Gowal et al. and Wang et al. have shown that generating additional training data using state-of-the-art diffusion models can considerably improve the robustness of adversarial training. In this work, we demonstrate that a similar approach can substantially improve deterministic certified defenses but also reveal notable differences in the scaling behavior between certified and empirical methods. In addition, we provide a list of recommendations to scale the robustness of certified training approaches. Our approach achieves state-of-the-art deterministic robustness certificates on CIFAR-10 for the
Red-teaming, or identifying prompts that elicit harmful responses, is a critical step in ensuring the safe and responsible deployment of lar… (voir plus)ge language models (LLMs). Developing effective protection against many modes of attack prompts requires discovering diverse attacks. Automated red-teaming typically uses reinforcement learning to fine-tune an attacker language model to generate prompts that elicit undesirable responses from a target LLM, as measured, for example, by an auxiliary toxicity classifier. We show that even with explicit regularization to favor novelty and diversity, existing approaches suffer from mode collapse or fail to generate effective attacks. As a flexible and probabilistically principled alternative, we propose to use GFlowNet fine-tuning, followed by a secondary smoothing phase, to train the attacker model to generate diverse and effective attack prompts. We find that the attacks generated by our method are effective against a wide range of target LLMs, both with and without safety tuning, and transfer well between target LLMs. Finally, we demonstrate that models safety-tuned using a dataset of red-teaming prompts generated by our method are robust to attacks from other RL-based red-teaming approaches.
Current research in adversarial robustness of LLMs focuses on discrete input manipulations in the natural language space, which can be direc… (voir plus)tly transferred to closed-source models. However, this approach neglects the steady progression of open-source models. As open-source models advance in capability, ensuring their safety also becomes increasingly imperative. Yet, attacks tailored to open-source LLMs that exploit full model access remain largely unexplored. We address this research gap and propose the embedding space attack, which directly attacks the continuous embedding representation of input tokens. We find that embedding space attacks circumvent model alignments and trigger harmful behaviors more efficiently than discrete attacks or model fine-tuning. Furthermore, we present a novel threat model in the context of unlearning and show that embedding space attacks can extract supposedly deleted information from unlearned LLMs across multiple datasets and models. Our findings highlight embedding space attacks as an important threat model in open-source LLMs. Trigger Warning: the appendix contains LLM-generated text with violence and harassment.
Current research in adversarial robustness of LLMs focuses on discrete input manipulations in the natural language space, which can be direc… (voir plus)tly transferred to closed-source models. However, this approach neglects the steady progression of open-source models. As open-source models advance in capability, ensuring their safety also becomes increasingly imperative. Yet, attacks tailored to open-source LLMs that exploit full model access remain largely unexplored. We address this research gap and propose the embedding space attack, which directly attacks the continuous embedding representation of input tokens. We find that embedding space attacks circumvent model alignments and trigger harmful behaviors more efficiently than discrete attacks or model fine-tuning. Furthermore, we present a novel threat model in the context of unlearning and show that embedding space attacks can extract supposedly deleted information from unlearned LLMs across multiple datasets and models. Our findings highlight embedding space attacks as an important threat model in open-source LLMs. Trigger Warning: the appendix contains LLM-generated text with violence and harassment.
Despite significant investment into safety training, large language models (LLMs) deployed in the real world still suffer from numerous vuln… (voir plus)erabilities. One perspective on LLM safety training is that it algorithmically forbids the model from answering toxic or harmful queries. To assess the effectiveness of safety training, in this work, we study forbidden tasks, i.e., tasks the model is designed to refuse to answer. Specifically, we investigate whether in-context learning (ICL) can be used to re-learn forbidden tasks despite the explicit fine-tuning of the model to refuse them. We first examine a toy example of refusing sentiment classification to demonstrate the problem. Then, we use ICL on a model fine-tuned to refuse to summarise made-up news articles. Finally, we investigate whether ICL can undo safety training, which could represent a major security risk. For the safety task, we look at Vicuna-7B, Starling-7B, and Llama2-7B. We show that the attack works out-of-the-box on Starling-7B and Vicuna-7B but fails on Llama2-7B. Finally, we propose an ICL attack that uses the chat template tokens like a prompt injection attack to achieve a better attack success rate on Vicuna-7B and Starling-7B. Trigger Warning: the appendix contains LLM-generated text with violence, suicide, and misinformation.
