Effective Red-Teaming of Policy-Adherent Agents

Task-oriented LLM-based agents are increasingly used in domains with strict policies, such as refund eligibility or cancellation rules. The challenge lies in ensuring that the agent consistently adheres to these rules and policies, appropriately refusing any request that would violate them, while still maintaining a helpful and natural interaction. This calls for the development of tailored design and evaluation methodologies to ensure agent resilience against malicious user behavior. We propose a novel threat model that focuses on adversarial users aiming to exploit policy-adherent agents for personal benefit. To address this, we present CRAFT, a multi-agent red-teaming system that leverages policy-aware persuasive strategies to undermine a policy-adherent agent in a customer-service scenario, outperforming conventional jailbreak methods such as DAN prompts, emotional manipulation, and coercive. Building upon the existing tau-bench benchmark, we introduce tau-break, a complementary benchmark designed to rigorously assess the agent's robustness against manipulative user behavior. Finally, we evaluate several straightforward yet effective defense strategies. While these measures provide some protection, they fall short, highlighting the need for stronger, research-driven safeguards to protect policy-adherent agents from adversarial attacks

Think Again! The Effect of Test-Time Compute on Preferences, Opinions, and Beliefs of Large Language Models

As Large Language Models (LLMs) become deeply integrated into human life and increasingly influence decision-making, it's crucial to evaluate whether and to what extent they exhibit subjective preferences, opinions, and beliefs. These tendencies may stem from biases within the models, which may shape their behavior, influence the advice and recommendations they offer to users, and potentially reinforce certain viewpoints. This paper presents the Preference, Opinion, and Belief survey (POBs), a benchmark developed to assess LLMs' subjective inclinations across societal, cultural, ethical, and personal domains. We applied our benchmark to evaluate leading open- and closed-source LLMs, measuring desired properties such as reliability, neutrality, and consistency. In addition, we investigated the effect of increasing the test-time compute, through reasoning and self-reflection mechanisms, on those metrics. While effective in other tasks, our results show that these mechanisms offer only limited gains in our domain. Furthermore, we reveal that newer model versions are becoming less consistent and more biased toward specific viewpoints, highlighting a blind spot and a concerning trend. POBS: https://ibm.github.io/POBS

Breaking ReAct Agents: Foot-in-the-Door Attack Will Get You In

Following the advancement of large language models (LLMs), the development of LLM-based autonomous agents has become increasingly prevalent. As a result, the need to understand the security vulnerabilities of these agents has become a critical task. We examine how ReAct agents can be exploited using a straightforward yet effective method we refer to as the foot-in-the-door attack. Our experiments show that indirect prompt injection attacks, prompted by harmless and unrelated requests (such as basic calculations) can significantly increase the likelihood of the agent performing subsequent malicious actions. Our results show that once a ReAct agents thought includes a specific tool or action, the likelihood of executing this tool in the subsequent steps increases significantly, as the agent seldom re-evaluates its actions. Consequently, even random, harmless requests can establish a foot-in-the-door, allowing an attacker to embed malicious instructions into the agents thought process, making it more susceptible to harmful directives. To mitigate this vulnerability, we propose implementing a simple reflection mechanism that prompts the agent to reassess the safety of its actions during execution, which can help reduce the success of such attacks.

Exploring Straightforward Conversational Red-Teaming

Large language models (LLMs) are increasingly used in business dialogue systems but they pose security and ethical risks. Multi-turn conversations, where context influences the model's behavior, can be exploited to produce undesired responses. In this paper, we examine the effectiveness of utilizing off-the-shelf LLMs in straightforward red-teaming approaches, where an attacker LLM aims to elicit undesired output from a target LLM, comparing both single-turn and conversational red-teaming tactics. Our experiments offer insights into various usage strategies that significantly affect their performance as red teamers. They suggest that off-the-shelf models can act as effective red teamers and even adjust their attack strategy based on past attempts, although their effectiveness decreases with greater alignment.

Can You Trust Your Metric? Automatic Concatenation-Based Tests for Metric Validity

Consider a scenario where a harmfulness detection metric is employed by a system to filter unsafe responses generated by a Large Language Model. When analyzing individual harmful and unethical prompt-response pairs, the metric correctly classifies each pair as highly unsafe, assigning the highest score. However, when these same prompts and responses are concatenated, the metric's decision flips, assigning the lowest possible score, thereby misclassifying the content as safe and allowing it to bypass the filter. In this study, we discovered that several harmfulness LLM-based metrics, including GPT-based, exhibit this decision-flipping phenomenon. Additionally, we found that even an advanced metric like GPT-4o is highly sensitive to input order. Specifically, it tends to classify responses as safe if the safe content appears first, regardless of any harmful content that follows, and vice versa. This work introduces automatic concatenation-based tests to assess the fundamental properties a valid metric should satisfy. We applied these tests in a model safety scenario to assess the reliability of harmfulness detection metrics, uncovering a number of inconsistencies.

From Zero to Hero: Cold-Start Anomaly Detection

When first deploying an anomaly detection system, e.g., to detect out-of-scope queries in chatbots, there are no observed data, making data-driven approaches ineffective. Zero-shot anomaly detection methods offer a solution to such "cold-start" cases, but unfortunately they are often not accurate enough. This paper studies the realistic but underexplored cold-start setting where an anomaly detection model is initialized using zero-shot guidance, but subsequently receives a small number of contaminated observations (namely, that may include anomalies). The goal is to make efficient use of both the zero-shot guidance and the observations. We propose ColdFusion, a method that effectively adapts the zero-shot anomaly detector to contaminated observations. To support future development of this new setting, we propose an evaluation suite consisting of evaluation protocols and metrics.

Detectors for Safe and Reliable LLMs: Implementations, Uses, and Limitations

Large language models (LLMs) are susceptible to a variety of risks, from non-faithful output to biased and toxic generations. Due to several limiting factors surrounding LLMs (training cost, API access, data availability, etc.), it may not always be feasible to impose direct safety constraints on a deployed model. Therefore, an efficient and reliable alternative is required. To this end, we present our ongoing efforts to create and deploy a library of detectors: compact and easy-to-build classification models that provide labels for various harms. In addition to the detectors themselves, we discuss a wide range of uses for these detector models - from acting as guardrails to enabling effective AI governance. We also deep dive into inherent challenges in their development and discuss future work aimed at making the detectors more reliable and broadening their scope.

Unveiling Safety Vulnerabilities of Large Language Models

As large language models become more prevalent, their possible harmful or inappropriate responses are a cause for concern. This paper introduces a unique dataset containing adversarial examples in the form of questions, which we call AttaQ, designed to provoke such harmful or inappropriate responses. We assess the efficacy of our dataset by analyzing the vulnerabilities of various models when subjected to it. Additionally, we introduce a novel automatic approach for identifying and naming vulnerable semantic regions - input semantic areas for which the model is likely to produce harmful outputs. This is achieved through the application of specialized clustering techniques that consider both the semantic similarity of the input attacks and the harmfulness of the model's responses. Automatically identifying vulnerable semantic regions enhances the evaluation of model weaknesses, facilitating targeted improvements to its safety mechanisms and overall reliability.

Characterizing how 'distributional' NLP corpora distance metrics are

A corpus of vector-embedded text documents has some empirical distribution. Given two corpora, we want to calculate a single metric of distance (e.g., Mauve, Frechet Inception) between them. We describe an abstract quality, called `distributionality', of such metrics. A non-distributional metric tends to use very local measurements, or uses global measurements in a way that does not fully reflect the distributions' true distance. For example, if individual pairwise nearest-neighbor distances are low, it may judge the two corpora to have low distance, even if their two distributions are in fact far from each other. A more distributional metric will, in contrast, better capture the distributions' overall distance. We quantify this quality by constructing a Known-Similarity Corpora set from two paraphrase corpora and calculating the distance between paired corpora from it. The distances' trend shape as set element separation increases should quantify the distributionality of the metric. We propose that Average Hausdorff Distance and energy distance between corpora are representative examples of non-distributional and distributional distance metrics, to which other metrics can be compared, to evaluate how distributional they are.

Measuring the Measuring Tools: An Automatic Evaluation of Semantic Metrics for Text Corpora

The ability to compare the semantic similarity between text corpora is important in a variety of natural language processing applications. However, standard methods for evaluating these metrics have yet to be established. We propose a set of automatic and interpretable measures for assessing the characteristics of corpus-level semantic similarity metrics, allowing sensible comparison of their behavior. We demonstrate the effectiveness of our evaluation measures in capturing fundamental characteristics by evaluating them on a collection of classical and state-of-the-art metrics. Our measures revealed that recently-developed metrics are becoming better in identifying semantic distributional mismatch while classical metrics are more sensitive to perturbations in the surface text levels.