In this study, we focus on two main tasks, the first for detecting legal violations within unstructured textual data, and the second for associating these violations with potentially affected individuals. We constructed two datasets using Large Language Models (LLMs) which were subsequently validated by domain expert annotators. Both tasks were designed specifically for the context of class-action cases. The experimental design incorporated fine-tuning models from the BERT family and open-source LLMs, and conducting few-shot experiments using closed-source LLMs. Our results, with an F1-score of 62.69\% (violation identification) and 81.02\% (associating victims), show that our datasets and setups can be used for both tasks. Finally, we publicly release the datasets and the code used for the experiments in order to advance further research in the area of legal natural language processing (NLP).

The advent of generative AI images has completely disrupted the art world. Distinguishing AI generated images from human art is a challenging problem whose impact is growing over time. A failure to address this problem allows bad actors to defraud individuals paying a premium for human art and companies whose stated policies forbid AI imagery. It is also critical for content owners to establish copyright, and for model trainers interested in curating training data in order to avoid potential model collapse. There are several different approaches to distinguishing human art from AI images, including classifiers trained by supervised learning, research tools targeting diffusion models, and identification by professional artists using their knowledge of artistic techniques. In this paper, we seek to understand how well these approaches can perform against today's modern generative models in both benign and adversarial settings. We curate real human art across 7 styles, generate matching images from 5 generative models, and apply 8 detectors (5 automated detectors and 3 different human groups including 180 crowdworkers, 4000+ professional artists, and 13 expert artists experienced at detecting AI). Both Hive and expert artists do very well, but make mistakes in different ways (Hive is weaker against adversarial perturbations while Expert artists produce higher false positives). We believe these weaknesses will remain as models continue to evolve, and use our data to demonstrate why a combined team of human and automated detectors provides the best combination of accuracy and robustness.

Planning and/or defending against decentralized info ops can be aided by computational simulations in lieu of ethically-fraught live experiments on social media. In this study, we introduce Diluvsion, an agent-based model for contested information propagation efforts on Twitter-like social media. The model emphasizes a user's belief in an opinion (stance) being impacted by the perception of potentially illusory popular support from constant incoming floods of indirect information, floods that can be cooperatively engineered in an uncoordinated manner by bots as they compete to spread their stances. Our model, which has been validated against real-world data, is an advancement over previous models because we account for engagement metrics in influencing stance adoption, non-social tie spreading of information, neutrality as a stance that can be spread, and themes that are analogous to media's framing effect and are symbiotic with respect to stance propagation. The strengths of the Diluvsion model are demonstrated in simulations of orthodox info ops, e.g., maximizing adoption of one stance; creating echo chambers; inducing polarization; and unorthodox info ops, e.g., simultaneous support of multiple stances as a Trojan horse tactic for the dissemination of a theme.

The conventional discourse on existential risks (x-risks) from AI typically focuses on abrupt, dire events caused by advanced AI systems, particularly those that might achieve or surpass human-level intelligence. These events have severe consequences that either lead to human extinction or irreversibly cripple human civilization to a point beyond recovery. This discourse, however, often neglects the serious possibility of AI x-risks manifesting incrementally through a series of smaller yet interconnected disruptions, gradually crossing critical thresholds over time. This paper contrasts the conventional "decisive AI x-risk hypothesis" with an "accumulative AI x-risk hypothesis." While the former envisions an overt AI takeover pathway, characterized by scenarios like uncontrollable superintelligence, the latter suggests a different causal pathway to existential catastrophes. This involves a gradual accumulation of critical AI-induced threats such as severe vulnerabilities and systemic erosion of econopolitical structures. The accumulative hypothesis suggests a boiling frog scenario where incremental AI risks slowly converge, undermining resilience until a triggering event results in irreversible collapse. Through systems analysis, this paper examines the distinct assumptions differentiating these two hypotheses. It is then argued that the accumulative view reconciles seemingly incompatible perspectives on AI risks. The implications of differentiating between these causal pathways -- the decisive and the accumulative -- for the governance of AI risks as well as long-term AI safety are discussed.

Automated red teaming holds substantial promise for uncovering and mitigating the risks associated with the malicious use of large language models (LLMs), yet the field lacks a standardized evaluation framework to rigorously assess new methods. To address this issue, we introduce HarmBench, a standardized evaluation framework for automated red teaming. We identify several desirable properties previously unaccounted for in red teaming evaluations and systematically design HarmBench to meet these criteria. Using HarmBench, we conduct a large-scale comparison of 18 red teaming methods and 33 target LLMs and defenses, yielding novel insights. We also introduce a highly efficient adversarial training method that greatly enhances LLM robustness across a wide range of attacks, demonstrating how HarmBench enables codevelopment of attacks and defenses. We open source HarmBench at https://github.com/centerforaisafety/HarmBench.

Large language models (LLMs) can pass explicit bias tests but still harbor implicit biases, similar to humans who endorse egalitarian beliefs yet exhibit subtle biases. Measuring such implicit biases can be a challenge: as LLMs become increasingly proprietary, it may not be possible to access their embeddings and apply existing bias measures; furthermore, implicit biases are primarily a concern if they affect the actual decisions that these systems make. We address both of these challenges by introducing two measures of bias inspired by psychology: LLM Implicit Association Test (IAT) Bias, which is a prompt-based method for revealing implicit bias; and LLM Decision Bias for detecting subtle discrimination in decision-making tasks. Using these measures, we found pervasive human-like stereotype biases in 6 LLMs across 4 social domains (race, gender, religion, health) and 21 categories (weapons, guilt, science, career among others). Our prompt-based measure of implicit bias correlates with embedding-based methods but better predicts downstream behaviors measured by LLM Decision Bias. This measure is based on asking the LLM to decide between individuals, motivated by psychological results indicating that relative not absolute evaluations are more related to implicit biases. Using prompt-based measures informed by psychology allows us to effectively expose nuanced biases and subtle discrimination in proprietary LLMs that do not show explicit bias on standard benchmarks.

This paper focuses on language change based on shifting social norms, in particular with regard to the debate on language and gender. It is a recurring argument in this debate that language develops "naturally" and that "severe interventions" - such as gender-inclusive language is often claimed to be - in the allegedly "organic" language system are inappropriate and even "dangerous". Such interventions are, however, not unprecedented. Socially motivated processes of language change are neither unusual nor new. We focus in our contribution on one important political-social space in Germany, the German Bundestag. Taking other struggles about language and gender in the plenaries of the Bundestag as a starting point, our article illustrates that language and gender has been a recurring issue in the German Bundestag since the 1980s. We demonstrate how this is reflected in linguistic practices of the Bundestag, by the use of a) designations for gays and lesbians; b) pair forms such as B\"urgerinnen und B\"urger (female and male citizens); and c) female forms of addresses and personal nouns ('Pr\"asidentin' in addition to 'Pr\"asident'). Lastly, we will discuss implications of these earlier language battles for the currently very heated debate about gender-inclusive language, especially regarding new forms with gender symbols like the asterisk or the colon (Lehrer*innen, Lehrer:innen; male*female teachers) which are intended to encompass all gender identities.

Machine learning algorithms may have disparate impacts on protected groups. To address this, we develop methods for Bayes-optimal fair classification, aiming to minimize classification error subject to given group fairness constraints. We introduce the notion of \emph{linear disparity measures}, which are linear functions of a probabilistic classifier; and \emph{bilinear disparity measures}, which are also linear in the group-wise regression functions. We show that several popular disparity measures -- the deviations from demographic parity, equality of opportunity, and predictive equality -- are bilinear. We find the form of Bayes-optimal fair classifiers under a single linear disparity measure, by uncovering a connection with the Neyman-Pearson lemma. For bilinear disparity measures, Bayes-optimal fair classifiers become group-wise thresholding rules. Our approach can also handle multiple fairness constraints (such as equalized odds), and the common scenario when the protected attribute cannot be used at the prediction phase. Leveraging our theoretical results, we design methods that learn fair Bayes-optimal classifiers under bilinear disparity constraints. Our methods cover three popular approaches to fairness-aware classification, via pre-processing (Fair Up- and Down-Sampling), in-processing (Fair Cost-Sensitive Classification) and post-processing (a Fair Plug-In Rule). Our methods control disparity directly while achieving near-optimal fairness-accuracy tradeoffs. We show empirically that our methods compare favorably to existing algorithms.

Efforts to align Large Language Models (LLMs) are mainly conducted via Reinforcement Learning from Human Feedback (RLHF) methods. However, RLHF encounters major challenges including training reward models, actor-critic engineering, and importantly, it requires access to LLM parameters. Here we introduce Aligner, a new efficient alignment paradigm that bypasses the whole RLHF process by learning the correctional residuals between the aligned and the unaligned answers. Our Aligner offers several key advantages. Firstly, it is an autoregressive seq2seq model that is trained on the query-answer-correction dataset via supervised learning; this offers a parameter-efficient alignment solution with minimal resources. Secondly, the Aligner facilitates weak-to-strong generalization; finetuning large pretrained models by Aligner's supervisory signals demonstrates strong performance boost. Thirdly, Aligner functions as a model-agnostic plug-and-play module, allowing for its direct application on different open-source and API-based models. Remarkably, Aligner-7B improves 11 different LLMs by 21.9% in helpfulness and 23.8% in harmlessness on average (GPT-4 by 17.5% and 26.9%). When finetuning (strong) Llama2-70B with (weak) Aligner-13B's supervision, we can improve Llama2 by 8.2% in helpfulness and 61.6% in harmlessness. See our dataset and code at https://aligner2024.github.io

Modern software has been an integral part of everyday activities in many disciplines and application contexts. Introducing intelligent automation by leveraging artificial intelligence (AI) led to break-throughs in many fields. The effectiveness of AI can be attributed to several factors, among which is the increasing availability of data. Regulations such as the general data protection regulation (GDPR) in the European Union (EU) are introduced to ensure the protection of personal data. Software systems that collect, process, or share personal data are subject to compliance with such regulations. Developing compliant software depends heavily on addressing legal requirements stipulated in applicable regulations, a central activity in the requirements engineering (RE) phase of the software development process. RE is concerned with specifying and maintaining requirements of a system-to-be, including legal requirements. Legal agreements which describe the policies organizations implement for processing personal data can provide an additional source to regulations for eliciting legal requirements. In this chapter, we explore a variety of methods for analyzing legal requirements and exemplify them on GDPR. Specifically, we describe possible alternatives for creating machine-analyzable representations from regulations, survey the existing automated means for enabling compliance verification against regulations, and further reflect on the current challenges of legal requirements analysis.