With LLM usage rapidly increasing, their vulnerability to jailbreaks that create harmful outputs are a major security risk. As new jailbreaking strategies emerge and models are changed by fine-tuning, continuous testing for security vulnerabilities is necessary. Existing Red Teaming methods fall short in cost efficiency, attack success rate, attack diversity, or extensibility as new attack types emerge. We address these challenges with Modular And Diverse Malicious Attack MiXtures (MAD-MAX) for Automated LLM Red Teaming. MAD-MAX uses automatic assignment of attack strategies into relevant attack clusters, chooses the most relevant clusters for a malicious goal, and then combines strategies from the selected clusters to achieve diverse novel attacks with high attack success rates. MAD-MAX further merges promising attacks together at each iteration of Red Teaming to boost performance and introduces a similarity filter to prune out similar attacks for increased cost efficiency. The MAD-MAX approach is designed to be easily extensible with newly discovered attack strategies and outperforms the prominent Red Teaming method Tree of Attacks with Pruning (TAP) significantly in terms of Attack Success Rate (ASR) and queries needed to achieve jailbreaks. MAD-MAX jailbreaks 97% of malicious goals in our benchmarks on GPT-4o and Gemini-Pro compared to TAP with 66%. MAD-MAX does so with only 10.9 average queries to the target LLM compared to TAP with 23.3. WARNING: This paper contains contents which are offensive in nature.

As large language models (LLMs) become integrated into everyday applications, ensuring their robustness and security is increasingly critical. In particular, LLMs can be manipulated into unsafe behaviour by prompts known as jailbreaks. The variety of jailbreak styles is growing, necessitating the use of external defences known as guardrails. While many jailbreak defences have been proposed, not all defences are able to handle new out-of-distribution attacks due to the narrow segment of jailbreaks used to align them. Moreover, the lack of systematisation around defences has created significant gaps in their practical application. In this work, we perform systematic benchmarking across 15 different defences, considering a broad swathe of malicious and benign datasets. We find that there is significant performance variation depending on the style of jailbreak a defence is subject to. Additionally, we show that based on current datasets available for evaluation, simple baselines can display competitive out-of-distribution performance compared to many state-of-the-art defences. Code is available at https://github.com/IBM/Adversarial-Prompt-Evaluation.

We introduce the Granite Guardian models, a suite of safeguards designed to provide risk detection for prompts and responses, enabling safe and responsible use in combination with any large language model (LLM). These models offer comprehensive coverage across multiple risk dimensions, including social bias, profanity, violence, sexual content, unethical behavior, jailbreaking, and hallucination-related risks such as context relevance, groundedness, and answer relevance for retrieval-augmented generation (RAG). Trained on a unique dataset combining human annotations from diverse sources and synthetic data, Granite Guardian models address risks typically overlooked by traditional risk detection models, such as jailbreaks and RAG-specific issues. With AUC scores of 0.871 and 0.854 on harmful content and RAG-hallucination-related benchmarks respectively, Granite Guardian is the most generalizable and competitive model available in the space. Released as open-source, Granite Guardian aims to promote responsible AI development across the community.

The proliferation of Large Language Models (LLMs) in diverse applications underscores the pressing need for robust security measures to thwart potential jailbreak attacks. These attacks exploit vulnerabilities within LLMs, endanger data integrity and user privacy. Guardrails serve as crucial protective mechanisms against such threats, but existing models often fall short in terms of both detection accuracy, and computational efficiency. This paper advocates for the significance of jailbreak attack prevention on LLMs, and emphasises the role of input guardrails in safeguarding these models. We introduce MoJE (Mixture of Jailbreak Expert), a novel guardrail architecture designed to surpass current limitations in existing state-of-the-art guardrails. By employing simple linguistic statistical techniques, MoJE excels in detecting jailbreak attacks while maintaining minimal computational overhead during model inference. Through rigorous experimentation, MoJE demonstrates superior performance capable of detecting 90% of the attacks without compromising benign prompts, enhancing LLMs security against jailbreak attacks.

As generative AI, particularly large language models (LLMs), become increasingly integrated into production applications, new attack surfaces and vulnerabilities emerge and put a focus on adversarial threats in natural language and multi-modal systems. Red-teaming has gained importance in proactively identifying weaknesses in these systems, while blue-teaming works to protect against such adversarial attacks. Despite growing academic interest in adversarial risks for generative AI, there is limited guidance tailored for practitioners to assess and mitigate these challenges in real-world environments. To address this, our contributions include: (1) a practical examination of red- and blue-teaming strategies for securing generative AI, (2) identification of key challenges and open questions in defense development and evaluation, and (3) the Attack Atlas, an intuitive framework that brings a practical approach to analyzing single-turn input attacks, placing it at the forefront for practitioners. This work aims to bridge the gap between academic insights and practical security measures for the protection of generative AI systems.

Current AI regulations require discarding sensitive features (e.g., gender, race, religion) in the algorithm's decision-making process to prevent unfair outcomes. However, even without sensitive features in the training set, algorithms can persist in discrimination. Indeed, when sensitive features are omitted (fairness under unawareness), they could be inferred through non-linear relations with the so called proxy features. In this work, we propose a way to reveal the potential hidden bias of a machine learning model that can persist even when sensitive features are discarded. This study shows that it is possible to unveil whether the black-box predictor is still biased by exploiting counterfactual reasoning. In detail, when the predictor provides a negative classification outcome, our approach first builds counterfactual examples for a discriminated user category to obtain a positive outcome. Then, the same counterfactual samples feed an external classifier (that targets a sensitive feature) that reveals whether the modifications to the user characteristics needed for a positive outcome moved the individual to the non-discriminated group. When this occurs, it could be a warning sign for discriminatory behavior in the decision process. Furthermore, we leverage the deviation of counterfactuals from the original sample to determine which features are proxies of specific sensitive information. Our experiments show that, even if the model is trained without sensitive features, it often suffers discriminatory biases.

The increasing application of Artificial Intelligence and Machine Learning models poses potential risks of unfair behavior and, in light of recent regulations, has attracted the attention of the research community. Several researchers focused on seeking new fairness definitions or developing approaches to identify biased predictions. However, none try to exploit the counterfactual space to this aim. In that direction, the methodology proposed in this work aims to unveil unfair model behaviors using counterfactual reasoning in the case of fairness under unawareness setting. A counterfactual version of equal opportunity named counterfactual fair opportunity is defined and two novel metrics that analyze the sensitive information of counterfactual samples are introduced. Experimental results on three different datasets show the efficacy of our methodologies and our metrics, disclosing the unfair behavior of classic machine learning and debiasing models.