Algorithmic fairness has conventionally adopted the mathematically convenient perspective of racial color-blindness (i.e., difference unaware treatment).However, we contend that in a range of important settings, group difference awareness matters.For example, differentiating between groups may be necessary in legal contexts (e.g., the U.S. compulsory draft applies to men but not women) and harm assessments (e.g., referring to girls as "terrorists" may be less harmful than referring to Muslim people as such).Thus, in contrast to most fairness work, we study fairness through the perspective of treating people differently -when it is contextually appropriate to.We first introduce an important distinction between descriptive (fact-based), normative (value-based), and correlation (association-based) benchmarks.This distinction is significant because each category requires separate interpretation and mitigation tailored to its specific characteristics.Then, we present a benchmark suite composed of eight different scenarios for a total of 16k questions that enables us to assess difference awareness.Finally, we show results across ten models that demonstrate difference awareness is a distinct dimension to fairness where existing bias mitigation strategies may backfire.

Regulatory agencies often operate with limited resources and rely on tips from the public to identify potential violations.However, processing these tips at scale presents significant operational challenges, as agencies must correctly identify and route relevant tips to the appropriate enforcement divisions.Through a case study, we demonstrate how advances in large language models can be utilized to support overburdened agencies with limited capacities.In partnership with the U.S. Environmental Protection Agency, we leverage previously unstudied citizen tips data from their "Report a Violation" system to develop an LLM-assisted pipeline for tip routing.Our approach filters out 80.5% of irrelevant tips and increases overall routing accuracy from 31.8% to 82.4% compared to the current routing system.At a time of increased focus on government efficiencies, our approach provides a constructive path forward by using technology to empower civil servants. * Equal contribution.

One of the emerging use cases of AI in law is for code simplification: streamlining, distilling, and simplifying complex statutory or regulatory language. One U.S. state has claimed to eliminate one third of its state code using AI. Yet we lack systematic evaluations of the accuracy, reliability, and risks of such approaches. We introduce LaborBench, a question-and-answer benchmark dataset designed to evaluate AI capabilities in this domain. We leverage a unique data source to create LaborBench: a dataset updated annually by teams of lawyers at the U.S. Department of Labor, who compile differences in unemployment insurance laws across 50 states for over 101 dimensions in a six-month process, culminating in a 200-page publication of tables. Inspired by our collaboration with one U.S. state to explore using large language models (LLMs) to simplify codes in this domain, where complexity is particularly acute, we transform the DOL publication into LaborBench. This provides a unique benchmark for AI capacity to conduct, distill, and extract realistic statutory and regulatory information. To assess the performance of retrieval augmented generation (RAG) approaches, we also compile StateCodes, a novel and comprehensive state statute and regulatory corpus of 8.7 GB, enabling much more systematic research into state codes. We then benchmark the performance of information retrieval and state-of-the-art large LLMs on this data and show that while these models are helpful as preliminary research for code simplification, the overall accuracy is far below the touted promises for LLMs as end-to-end pipelines for regulatory simplification.1

Each year thousands of bills are enacted into law by Congress, state legislatures, and municipalities. As new laws are enacted, legal codes have grown in length and complexity to such an extent that, in many domains, even the government lacks systematic knowledge of what the law is. To address the problem, we develop the Statutory Research Assistant (STARA), an automated system capable of performing accurate “statutory surveys”—compilations of all provisions relevant to a particular legal issue. We validate STARA’s accuracy on three existing human-compiled surveys, showing that it can reproduce them with high fidelity and in many cases identify hundreds of previously undiscovered provisions. We show that STARA’s domain-specific architecture yields substantial improvements in accuracy over off-the-shelf language models with considerable implications for legal reform, academic research, and transparency. We make STARA’s compiled statutory surveys available to the public and the tool available to researchers upon request.

AI audits play a critical role in AI accountability and safety. They are particularly salient in anti-discrimination law. Several areas of anti-discrimination law implicate what is known as the "less discriminatory alternative" (LDA) requirement, under which a protocol is defensible if no less discriminatory model that achieves comparable performance can be found with reasonable effort. Notably, the burden of proving an LDA exists typically falls on the claimant (the party alleging discrimination). This creates a significant hurdle in AI cases, as the claimant would seemingly need to train a less discriminatory yet high-performing model, a task requiring resources and expertise beyond most litigants. Moreover, developers often restrict access to their models and data as trade secrets, hindering replicability. In this work, we present a procedure enabling claimants to determine if an LDA exists, even when they have limited compute, data, and model access. To illustrate our approach, we focus on the setting in which fairness is given by demographic parity and performance by binary cross-entropy loss. As our main result, we provide a novel closed-form upper bound for the loss-fairness Pareto frontier (PF). This expression is powerful because the claimant can use it to fit the PF in the ''low-resource regime," then extrapolate the PF that applies to the (large) model being contested, all without training a single large model. The expression thus serves as a scaling law for loss-fairness PFs. To use this scaling law, the claimant would require a small subsample of the train/test data. Then, for a given compute budget, the claimant can fit the context-specific PF by training as few as 7 (small) models. We stress test our main result in simulations, finding that our scaling law applies even when the exact conditions of our theory do not hold.

"Machine unlearning" is a popular proposed solution for mitigating the existence of content in an AI model that is problematic for legal or moral reasons, including privacy, copyright, safety, and more. For example, unlearning is often invoked as a solution for removing the effects of specific information from a generative-AI model's parameters, e.g., a particular individual's personal data or the inclusion of copyrighted content 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 ML researchers and policymakers to think rigorously about these challenges, identifying several mismatches between the goals of unlearning and feasible implementations. These mismatches explain why unlearning is not a general-purpose solution for circumscribing generative-AI model behavior in service of broader positive impact.

We study the problem of efficiently estimating the mean of a $k$-class random variable, $Y$, using a limited number of labels, $N$, in settings where the analyst has access to auxiliary information (i.e.: covariates) $X$ that may be informative about $Y$. We propose an active learning algorithm ("PartiBandits") to estimate $\mathbb{E}[Y]$. The algorithm yields an estimate, $\widehat{\mu}_{\text{PB}}$, such that $\left( \widehat{\mu}_{\text{PB}} - \mathbb{E}[Y]\right)^2$ is $\tilde{\mathcal{O}}\left( \frac{\nu + \exp(c \cdot (-N/\log(N))) }{N} \right)$, where $c > 0$ is a constant and $\nu$ is the risk of the Bayes-optimal classifier. PartiBandits is essentially a two-stage algorithm. In the first stage, it learns a partition of the unlabeled data that shrinks the average conditional variance of $Y$. In the second stage it uses a UCB-style subroutine ("WarmStart-UCB") to request labels from each stratum round-by-round. Both the main algorithm's and the subroutine's convergence rates are minimax optimal in classical settings. PartiBandits bridges the UCB and disagreement-based approaches to active learning despite these two approaches being designed to tackle very different tasks. We illustrate our methods through simulation using nationwide electronic health records. Our methods can be implemented using the \textbf{PartiBandits} package in R.

AI agents have the potential to significantly alter the cybersecurity landscape. Here, we introduce the first framework to capture offensive and defensive cyber-capabilities in evolving real-world systems. Instantiating this framework with BountyBench, we set up 25 systems with complex, real-world codebases. To capture the vulnerability lifecycle, we define three task types: Detect (detecting a new vulnerability), Exploit (exploiting a given vulnerability), and Patch (patching a given vulnerability). For Detect, we construct a new success indicator, which is general across vulnerability types and provides localized evaluation. We manually set up the environment for each system, including installing packages, setting up server(s), and hydrating database(s). We add 40 bug bounties, which are vulnerabilities with monetary awards from \\$10 to \\$30,485, covering 9 of the OWASP Top 10 Risks. To modulate task difficulty, we devise a new strategy based on information to guide detection, interpolating from identifying a zero day to exploiting a given vulnerability. We evaluate 10 agents: Claude Code, OpenAI Codex CLI with o3-high and o4-mini, and custom agents with o3-high, GPT-4.1, Gemini 2.5 Pro Preview, Claude 3.7 Sonnet Thinking, Qwen3 235B A22B, Llama 4 Maverick, and DeepSeek-R1. Given up to three attempts, the top-performing agents are OpenAI Codex CLI: o3-high (12.5% on Detect, mapping to \\$3,720; 90% on Patch, mapping to \\$14,152), Custom Agent with Claude 3.7 Sonnet Thinking (67.5% on Exploit), and OpenAI Codex CLI: o4-mini (90% on Patch, mapping to \\$14,422). OpenAI Codex CLI: o3-high, OpenAI Codex CLI: o4-mini, and Claude Code are more capable at defense, achieving higher Patch scores of 90%, 90%, and 87.5%, compared to Exploit scores of 47.5%, 32.5%, and 57.5% respectively; while the custom agents are relatively balanced between offense and defense, achieving Exploit scores of 17.5-67.5% and Patch scores of 25-60%.

Large, high-quality datasets are crucial for training Large Language Models (LLMs).However, so far, few datasets are available for specialized critical domains such as law and the available ones are often small and only in English.To fill this gap, we curate and release MULTILEGALPILE, a 689GB corpus in 24 languages from 17 jurisdictions.MULTILE-GALPILE includes diverse legal data sources and allows for pretraining NLP models under fair use, with most of the dataset licensed very permissively.We pretrain two RoBERTa models and one Longformer multilingually, and 24 monolingual models on each of the languagespecific subsets and evaluate them on LEX-TREME.Additionally, we evaluate the English and multilingual models on LexGLUE.Our multilingual models set a new SotA on LEX-TREME and our English models on LexGLUE.We release the dataset, trained models, and all code under the most open licenses possible.

Static word embeddings are ubiquitous in computational social science applications and contribute to practical decision-making in a variety of fields including law and healthcare.However, assessing the statistical uncertainty in downstream conclusions drawn from word embedding statistics has remained challenging.When using only point estimates for embeddings, researchers have no streamlined way of assessing the degree to which their model selection criteria or scientific conclusions are subject to noise due to sparsity in the underlying data used to generate the embeddings.We introduce a method to obtain approximate, easy-to-use, and scalable reconstruction error variance estimates for GloVe (Pennington et al., 2014), one of the most widely used word embedding models, using an analytical approximation to a multivariate normal model.To demonstrate the value of embeddings with variance (GloVe-V), we illustrate how our approach enables principled hypothesis testing in core word embedding tasks, such as comparing the similarity between different word pairs in vector space, assessing the performance of different models, and analyzing the relative degree of ethnic or gender bias in a corpus using different word lists.

Foundation models are powerful technologies: how they are released publicly directly shapes their societal impact. In this position paper, we focus on *open* foundation models, defined here as those with broadly available model weights (e.g., Llama 3, Stable Diffusion XL). We identify five distinctive properties (e.g., greater customizability, poor monitoring) that mediate their benefits and risks. Open foundation models present significant benefits, with some caveats, that span innovation, competition, the distribution of decision-making power, and transparency. To understand their risks of misuse, we design a risk assessment framework for analyzing their *marginal risk*. Across several misuse vectors (e.g., cyberattacks, bioweapons), we find that current research is insufficient to effectively characterize the marginal risk of open foundation models relative to pre-existing technologies. The framework helps explain why the marginal risk is low in some cases, clarifies disagreements about misuse risks by revealing that past work has focused on different subsets of the framework with different assumptions, and articulates a way forward for more constructive debate. Overall, our work helps support a more grounded assessment of the societal impact of open foundation models by outlining what research is needed to empirically validate their theoretical benefits and risks.

We introduce a new setting, optimize-and-estimate structured bandits. Here, a policy must select a batch of arms, each characterized by its own context, that would allow it to both maximize reward and maintain an accurate (ideally unbiased) population estimate of the reward. This setting is inherent to many public and private sector applications and often requires handling delayed feedback, small data, and distribution shifts. We demonstrate its importance on real data from the United States Internal Revenue Service (IRS). The IRS performs yearly audits of the tax base. Two of its most important objectives are to identify suspected misreporting and to estimate the "tax gap" -- the global difference between the amount paid and true amount owed. Based on a unique collaboration with the IRS, we cast these two processes as a unified optimize-and-estimate structured bandit. We analyze optimize-and-estimate approaches to the IRS problem and propose a novel mechanism for unbiased population estimation that achieves rewards comparable to baseline approaches. This approach has the potential to improve audit efficacy, while maintaining policy-relevant estimates of the tax gap. This has important social consequences given that the current tax gap is estimated at nearly half a trillion dollars. We suggest that this problem setting is fertile ground for further research and we highlight its interesting challenges. The results of this and related research are currently being incorporated into the continual improvement of the IRS audit selection methods.

Entropy regularization is known to improve exploration in sequential decision-making problems. We show that this same mechanism can also lead to nearly unbiased and lower-variance estimates of the mean reward in the optimize-and-estimate structured bandit setting. Mean reward estimation (i.e., population estimation) tasks have recently been shown to be essential for public policy settings where legal constraints often require precise estimates of population metrics. We show that leveraging entropy and KL divergence can yield a better trade-off between reward and estimator variance than existing baselines, all while remaining nearly unbiased. These properties of entropy regularization illustrate an exciting potential for bringing together the optimal exploration and estimation literature.

Can government govern artificial intelligence (AI)? One of the central questions of AI governance surrounds state capacity, namely whether government has the ability to accomplish its policy goals. We study this question by assessing how well the U.S. federal government has implemented three binding laws around AI governance: two executive orders—concerning trustworthy AI in the public sector (E.O. 13,960) and AI leadership (E.O. 13,859)—and the AI in Government Act. We conduct the first systematic empirical assessment of the implementation status of these three laws, which have each been described as central to US AI innovation. First, we track, through extensive research, line-level adoption of each mandated action. Based on publicly available information, we find that fewer than 40 percent of 45 legal requirements could be verified as having been implemented. Second, we research the specific implementation of transparency requirements at up to 220 federal agencies. We find that nearly half of agencies failed to publicly issue AI use case inventories—even when these agencies have demonstrable use cases of machine learning. Even when agencies have complied with these requirements, efforts are inconsistent. Our work highlights the weakness of U.S. state capacity to carry out AI governance mandates and we discuss implications for how to address bureaucratic capacity challenges.

The advent of large language models (LLMs) and their adoption by the legal community has given rise to the question: what types of legal reasoning can LLMs perform? To enable greater study of this question, we present LegalBench: a collaboratively constructed legal reasoning benchmark consisting of 162 tasks covering six different types of legal reasoning. LegalBench was built through an interdisciplinary process, in which we collected tasks designed and hand-crafted by legal professionals. Because these subject matter experts took a leading role in construction, tasks either measure legal reasoning capabilities that are practically useful, or measure reasoning skills that lawyers find interesting. To enable cross-disciplinary conversations about LLMs in the law, we additionally show how popular legal frameworks for describing legal reasoning—which distinguish between its many forms—correspond to LegalBench tasks, thus giving lawyers and LLM developers a common vocabulary. This paper describes LegalBench, presents an empirical evaluation of 20 open-source and commercial LLMs, and illustrates the types of research explorations LegalBench enables.

Much attention has focused on algorithmic audits and impact assessments to hold developers and users of algorithmic systems accountable. But existing algorithmic accountability policy approaches have neglected the lessons from non-algorithmic domains: notably, the importance of third parties. Our paper synthesizes lessons from other fields on how to craft effective systems of external oversight for algorithmic deployments. First, we discuss the challenges of third party oversight in the current AI landscape. Second, we survey audit systems across domains - e.g., financial, environmental, and health regulation - and show that the institutional design of such audits are far from monolithic. Finally, we survey the evidence base around these design components and spell out the implications for algorithmic auditing. We conclude that the turn toward audits alone is unlikely to achieve actual algorithmic accountability, and sustained focus on institutional design will be required for meaningful third party involvement.

This paper introduces a new, highly consequential setting for the use of computer vision for environmental sustainability. Concentrated Animal Feeding Operations (CAFOs) (aka intensive livestock farms or "factory farms") produce significant manure and pollution. Dumping manure in the winter months poses significant environmental risks and violates environmental law in many states. Yet the federal Environmental Protection Agency (EPA) and state agencies have relied primarily on self-reporting to monitor such instances of "land application." Our paper makes four contributions. First, we introduce the environmental, policy, and agricultural setting of CAFOs and land application. Second, we provide a new dataset of high-cadence (daily to weekly) 3m/pixel satellite imagery from 2018-20 for 330 CAFOs in Wisconsin with hand labeled instances of land application (n=57,697). Third, we develop an object detection model to predict land application and a system to perform inference in near real-time. We show that this system effectively appears to detect land application (PR AUC = 0.93) and we uncover several outlier facilities which appear to apply regularly and excessively. Last, we estimate the population prevalence of land application events in Winter 2021/22. We show that the prevalence of land application is much higher than what is self-reported by facilities. The system can be used by environmental regulators and interest groups, one of which piloted field visits based on this system this past winter. Overall, our application demonstrates the potential for AI-based computer vision systems to solve major problems in environmental compliance with near-daily imagery.

One concern with the rise of large language models lies with their potential for significant harm, particularly from pretraining on biased, obscene, copyrighted, and private information. Emerging ethical approaches have attempted to filter pretraining material, but such approaches have been ad hoc and failed to take context into account. We offer an approach to filtering grounded in law, which has directly addressed the tradeoffs in filtering material. First, we gather and make available the Pile of Law, a ~256GB (and growing) dataset of open-source English-language legal and administrative data, covering court opinions, contracts, administrative rules, and legislative records. Pretraining on the Pile of Law may help with legal tasks that have the promise to improve access to justice. Second, we distill the legal norms that governments have developed to constrain the inclusion of toxic or private content into actionable lessons for researchers and discuss how our dataset reflects these norms. Third, we show how the Pile of Law offers researchers the opportunity to learn such filtering rules directly from the data, providing an exciting new research direction in model-based processing.

Longitudinal studies are vital to understanding dynamic changes of the planet, but labels (e.g., buildings, facilities, roads) are often available only for a single point in time. We propose a general model, Temporal Cluster Matching (TCM), for detecting building changes in time series of remotely sensed imagery when footprint labels are observed only once. The intuition behind the model is that the relationship between spectral values inside and outside of building’s footprint will change when a building is constructed (or demolished). For instance, in rural settings, the pre-construction area may look similar to the surrounding environment until the building is constructed. Similarly, in urban settings, the pre-construction areas will look different from the surrounding environment until construction. We further propose a heuristic method for selecting the parameters of our model which allows it to be applied in novel settings without requiring data labeling efforts (to fit the parameters). We apply our model over a dataset of poultry barns from 2016/2017 high-resolution aerial imagery in the Delmarva Peninsula and a dataset of solar farms from a 2020 mosaic of Sentinel 2 imagery in India. Our results show that our model performs as well when fit using the proposed heuristic as it does when fit with labeled data, and further, that supervised versions of our model perform the best among all the baselines we test against. Finally, we show that our proposed approach can act as an effective data augmentation strategy – it enables researchers to augment existing structure footprint labels along the time dimension and thus use imagery from multiple points in time to train deep learning models. We show that this improves the spatial generalization of such models when evaluated on the same change detection task.

Lawyers and judges spend a large amount of time researching the proper legal authority to cite while drafting decisions. In this paper, we develop a citation recommendation tool that can help improve efficiency in the process of opinion drafting. We train four types of machine learning models, including a citation-list based method (collaborative filtering) and three context-based methods (text similarity, BiLSTM and RoBERTa classifiers). Our experiments show that leveraging local textual context improves recommendation, and that deep neural models achieve decent performance. We show that non-deep text-based methods benefit from access to structured case metadata, but deep models only benefit from such access when predicting from context of insufficient length. We also find that, even after extensive training, RoBERTa does not outperform a recurrent neural model, despite its benefits of pretraining. Our behavior analysis of the RoBERTa model further shows that predictive performance is stable across time and citation classes.

While self-supervised learning has made rapid advances in natural language processing, it remains unclear when researchers should engage in resource-intensive domain-specific pretraining (domain pretraining). The law, puzzlingly, has yielded few documented instances of substantial gains to domain pretraining in spite of the fact that legal language is widely seen to be unique. We hypothesize that these existing results stem from the fact that existing legal NLP tasks are too easy and fail to meet conditions for when domain pretraining can help. To address this, we first present CaseHOLD (Case Holdings On Legal Decisions), a new dataset comprised of over 53,000+ multiple choice questions to identify the relevant holding of a cited case. This dataset presents a fundamental task to lawyers and is both legally meaningful and difficult from an NLP perspective (F1 of 0.4 with a BiLSTM baseline). Second, we assess performance gains on CaseHOLD and existing legal NLP datasets. While a Transformer architecture (BERT) pretrained on a general corpus (Google Books and Wikipedia) improves performance, domain pretraining (on a corpus of ≈3.5M decisions across all courts in the U.S. that is larger than BERT's) with a custom legal vocabulary exhibits the most substantial performance gains with CaseHOLD (gain of 7.2% on F1, representing a 12% improvement on BERT) and consistent performance gains across two other legal tasks. Third, we show that domain pretraining may be warranted when the task exhibits sufficient similarity to the pretraining corpus: the level of performance increase in three legal tasks was directly tied to the domain specificity of the task. Our findings inform when researchers should engage in resource-intensive pretraining and show that Transformer-based architectures, too, learn embeddings suggestive of distinct legal language.

Social media provides the government with novel methods to improve regulation. One leading case has been the use of Yelp reviews to target food safety inspections. While previous research on data from Seattle finds that Yelp reviews can predict unhygienic establishments, we provide a more cautionary perspective. First, we show that prior results are sensitive to what we call “Extreme Imbalanced Sampling”: extreme because the dataset was restricted from roughly 13k inspections to a sample of only 612 inspections with only extremely high or low inspection scores, and imbalanced by not accounting for class imbalance in the population. We show that extreme imbalanced sampling is responsible for claims about the power of Yelp information in the original classification setup. Second, a re-analysis that utilizes the full dataset of 13k inspections and models the full inspection score (regression instead of classification) shows that (a) Yelp information has lower predictive power than prior inspection history and (b) Yelp reviews do not significantly improve predictions, given existing information about restaurants and inspection history. Contrary to prior claims, Yelp reviews do not appear to aid regulatory targeting. Third, this case study highlights critical issues when using social media for predictive models in governance and corroborates recent calls for greater transparency and reproducibility in machine learning.