The EU Artificial Intelligence Act (AIA) establishes different legal principles for different types of AI systems. While prior work has sought to clarify some of these principles, little attention has been paid to robustness and cybersecurity. This paper aims to fill this gap. We identify legal challenges and shortcomings in provisions related to robustness and cybersecurity for high-risk AI systems (Art. 15 AIA) and general-purpose AI models (Art. 55 AIA). We show that robustness and cybersecurity demand resilience against performance disruptions. Furthermore, we assess potential challenges in implementing these provisions in light of recent advancements in the machine learning (ML) literature. Our analysis informs efforts to develop harmonized standards, guidelines by the European Commission, as well as benchmarks and measurement methodologies under Art. 15(2) AIA. With this, we seek to bridge the gap between legal terminology and ML research, fostering a better alignment between research and implementation efforts.

Linear representation hypothesis posits that high-level concepts are encoded as linear directions in the representation spaces of LLMs. Park et al. (2024) formalize this notion by unifying multiple interpretations of linear representation, such as 1-dimensional subspace representation and interventions, using a causal inner product. However, their framework relies on single-token counterfactual pairs and cannot handle ambiguous contrasting pairs, limiting its applicability to complex or context-dependent concepts. We introduce a new notion of binary concepts as unit vectors in a canonical representation space, and utilize LLMs' (neural) activation differences along with maximum likelihood estimation (MLE) to compute concept directions (i.e., steering vectors). Our method, Sum of Activation-base Normalized Difference (SAND), formalizes the use of activation differences modeled as samples from a von Mises-Fisher (vMF) distribution, providing a principled approach to derive concept directions. We extend the applicability of Park et al. (2024) by eliminating the dependency on unembedding representations and single-token pairs. Through experiments with LLaMA models across diverse concepts and benchmarks, we demonstrate that our lightweight approach offers greater flexibility, superior performance in activation engineering tasks like monitoring and manipulation.

Building on related concepts, like, decentralized identifiers (DIDs), proof of personhood, anonymous credentials, personhood credentials (PHCs) emerged as an alternative approach, enabling individuals to verify to digital service providers that they are a person without disclosing additional information. However, new technologies might introduce some friction due to users misunderstandings and mismatched expectations. Despite their growing importance, limited research has been done on users perceptions and preferences regarding PHCs. To address this gap, we conducted competitive analysis, and semi-structured online user interviews with 23 participants from US and EU to provide concrete design recommendations for PHCs that incorporate user needs, adoption rules, and preferences. Our study -- (a)surfaces how people reason about unknown privacy and security guarantees of PHCs compared to current verification methods -- (b) presents the impact of several factors on how people would like to onboard and manage PHCs, including, trusted issuers (e.g. gov), ground truth data to issue PHC (e.g biometrics, physical id), and issuance system (e.g. centralized vs decentralized). In a think-aloud conceptual design session, participants recommended -- conceptualized design, such as periodic biometrics verification, time-bound credentials, visually interactive human-check, and supervision of government for issuance system. We propose actionable designs reflecting users preferences.

Vision-Language Models (VLMs) are increasingly deployed in public sector missions, necessitating robust evaluation of their safety and vulnerability to adversarial attacks. This paper introduces a novel framework to quantify adversarial risks in VLMs. We analyze model performance under Gaussian, salt-and-pepper, and uniform noise, identifying misclassification thresholds and deriving composite noise patches and saliency patterns that highlight vulnerable regions. These patterns are compared against the Fast Gradient Sign Method (FGSM) to assess their adversarial effectiveness. We propose a new Vulnerability Score that combines the impact of random noise and adversarial attacks, providing a comprehensive metric for evaluating model robustness.

The recent application of LLMs to the legal field has spurred the creation of benchmarks across various jurisdictions and languages. However, no benchmark has yet been specifically designed for the Portuguese legal system. In this work, we present LegalBench.PT, the first comprehensive legal benchmark covering key areas of Portuguese law. To develop LegalBench.PT, we first collect long-form questions and answers from real law exams, and then use GPT-4o to convert them into multiple-choice, true/false, and matching formats. Once generated, the questions are filtered and processed to improve the quality of the dataset. To ensure accuracy and relevance, we validate our approach by having a legal professional review a sample of the generated questions. Although the questions are synthetically generated, we show that their basis in human-created exams and our rigorous filtering and processing methods applied result in a reliable benchmark for assessing LLMs' legal knowledge and reasoning abilities. Finally, we evaluate the performance of leading LLMs on LegalBench.PT and investigate potential biases in GPT-4o's responses. We also assess the performance of Portuguese lawyers on a sample of questions to establish a baseline for model comparison and validate the benchmark.

Companies across all economic sectors continue to deploy large language models at a rapid pace. Reinforcement learning is experiencing a resurgence of interest due to its association with the fine-tuning of language models from human feedback. Tool-chain language models control task-specific agents; if the converse has not already appeared, it soon will. In this paper, we present what we believe is the first investigation of an intelligent trading agent based on continuous deep reinforcement learning that also controls a large language model with which it can post to a social media feed observed by other traders. We empirically investigate the performance and impact of such an agent in a simulated financial market, finding that it learns to optimize its total reward, and thereby augment its profit, by manipulating the sentiment of the posts it produces. The paper concludes with discussion, limitations, and suggestions for future work.

In strategic multi-agent sequential interactions, detecting dynamic coalition structures is crucial for understanding how self-interested agents coordinate to influence outcomes. However, natural-language-based interactions introduce unique challenges to coalition detection due to ambiguity over intents and difficulty in modeling players' subjective perspectives. We propose a new method that leverages recent advancements in large language models and game theory to predict dynamic multilateral coalition formation in Diplomacy, a strategic multi-agent game where agents negotiate coalitions using natural language. The method consists of two stages. The first stage extracts the set of agreements discussed by two agents in their private dialogue, by combining a parsing-based filtering function with a fine-tuned language model trained to predict player intents. In the second stage, we define a new metric using the concept of subjective rationalizability from hypergame theory to evaluate the expected value of an agreement for each player. We then compute this metric for each agreement identified in the first stage by assessing the strategic value of the agreement for both players and taking into account the subjective belief of one player that the second player would honor the agreement. We demonstrate that our method effectively detects potential coalition structures in online Diplomacy gameplay by assigning high values to agreements likely to be honored and low values to those likely to be violated. The proposed method provides foundational insights into coalition formation in multi-agent environments with language-based negotiation and offers key directions for future research on the analysis of complex natural language-based interactions between agents.

This tension in assumptions is readily apparent in standard human-AI alignment methods--such as reinforcement learning from human feedback (RLHF) [6, 7, 8] and direct preference optimization (DPO) [9]--which assume a single reward function captures the interests of the entire population. We examine the limits of the preference homogeneity assumption when individuals belong to user types, each characterized by a specific reward function. Recent work has shown that in this setting, the homogeneity assumption can lead to unexpected behavior [10, 11, 12]. One challenge is that, under this assumption, learning from human preferences becomes unrealizable, as a single reward function cannot capture the complexity of population preferences with multiple reward functions [13, 14]. Both RLHF and DPO rely on maximum likelihood estimation (MLE) to optimize the reward or policy. Unrealizability implies their likelihood functions cannot fully represent the underlying preference data distribution, resulting in a nontrivial optimal MLE solution. From another perspective, learning a universal reward or policy from a heterogeneous population inherently involves an aggregation of diverse interests, and this aggregation is nontrivial. In the quest for a single policy that accommodates a heterogeneous population with multiple user types, we show that the only universal reward yielding a well-defined alignment problem is an affine Equal contribution Work done while visiting Harvard Equal advising 1 arXiv:2502.16320v1

Generative AI (GenAI) is increasingly used in survey contexts to simulate human preferences. While many research endeavors evaluate the quality of synthetic GenAI data by comparing model-generated responses to gold-standard survey results, fundamental questions about the validity and reliability of using LLMs as substitutes for human respondents remain. Our study provides a technical analysis of how demographic attributes and prompt variations influence latent opinion mappings in large language models (LLMs) and evaluates their suitability for survey-based predictions. Using 14 different models, we find that LLM-generated data fails to replicate the variance observed in real-world human responses, particularly across demographic subgroups. In the political space, persona-to-party mappings exhibit limited differentiation, resulting in synthetic data that lacks the nuanced distribution of opinions found in survey data. Moreover, we show that prompt sensitivity can significantly alter outputs for some models, further undermining the stability and predictiveness of LLM-based simulations. As a key contribution, we adapt a probe-based methodology that reveals how LLMs encode political affiliations in their latent space, exposing the systematic distortions introduced by these models. Our findings highlight critical limitations in AI-generated survey data, urging caution in its use for public opinion research, social science experimentation, and computational behavioral modeling.

Natural Language Processing research is increasingly reliant on large scale data and computational power. Many achievements in the past decade resulted from collaborations with the tech industry. But an increasing entanglement of academic research and industry interests leads to conflicts of interest. We assessed published NLP research from 2000-2024 and labeled author affiliations as academic or industry-affiliated to measure conflicts of interest. Overall 27.65% of the papers contained at least one industry-affiliated author. That figure increased substantially with more than 1 in 3 papers having a conflict of interest in 2024. We identify top-tier venues (ACL, EMNLP) as main drivers for that effect. The paper closes with a discussion and a simple, concrete suggestion for the future.