Agent Benchmarks Fail Public Sector Requirements

Deploying Large Language Model-based agents (LLM agents) in the public sector requires assuring that they meet the stringent legal, procedural, and structural requirements of public-sector institutions. Practitioners and researchers often turn to benchmarks for such assessments. However, it remains unclear what criteria benchmarks must meet to ensure they adequately reflect public-sector requirements, or how many existing benchmarks do so. In this paper, we first define such criteria based on a first-principles survey of public administration literature: benchmarks must be \emph{process-based}, \emph{realistic}, \emph{public-sector-specific} and report \emph{metrics} that reflect the unique requirements of the public sector. We analyse more than 1,300 benchmark papers for these criteria using an expert-validated LLM-assisted pipeline. Our results show that no single benchmark meets all of the criteria. Our findings provide a call to action for both researchers to develop public sector-relevant benchmarks and for public-sector officials to apply these criteria when evaluating their own agentic use cases.

One Persona, Many Cues, Different Results: How Sociodemographic Cues Impact LLM Personalization

Personalization of LLMs by sociodemographic subgroup often improves user experience, but can also introduce or amplify biases and unfair outcomes across groups. Prior work has employed so-called personas, sociodemographic user attributes conveyed to a model, to study bias in LLMs by relying on a single cue to prompt a persona, such as user names or explicit attribute mentions. This disregards LLM sensitivity to prompt variations (robustness) and the rarity of some cues in real interactions (external validity). We compare six commonly used persona cues across seven open and proprietary LLMs on four writing and advice tasks. While cues are overall highly correlated, they produce substantial variance in responses across personas. We therefore caution against claims from a single persona cue and recommend future personalization research to evaluate multiple externally valid cues.

Who Evaluates AI's Social Impacts? Mapping Coverage and Gaps in First and Third Party Evaluations

Foundation models are increasingly central to high-stakes AI systems, and governance frameworks now depend on evaluations to assess their risks and capabilities. Although general capability evaluations are widespread, social impact assessments covering bias, fairness, privacy, environmental costs, and labor practices remain uneven across the AI ecosystem. To characterize this landscape, we conduct the first comprehensive analysis of both first-party and third-party social impact evaluation reporting across a wide range of model developers. Our study examines 186 first-party release reports and 183 post-release evaluation sources, and complements this quantitative analysis with interviews of model developers. We find a clear division of evaluation labor: first-party reporting is sparse, often superficial, and has declined over time in key areas such as environmental impact and bias, while third-party evaluators including academic researchers, nonprofits, and independent organizations provide broader and more rigorous coverage of bias, harmful content, and performance disparities. However, this complementarity has limits. Only model developers can authoritatively report on data provenance, content moderation labor, financial costs, and training infrastructure, yet interviews reveal that these disclosures are often deprioritized unless tied to product adoption or regulatory compliance. Our findings indicate that current evaluation practices leave major gaps in assessing AI's societal impacts, highlighting the urgent need for policies that promote developer transparency, strengthen independent evaluation ecosystems, and create shared infrastructure to aggregate and compare third-party evaluations in a consistent and accessible way.

Oversight Structures for Agentic AI in Public-Sector Organizations

This paper finds that the introduction of agentic AI systems intensifies existing challenges to traditional public sector oversight mechanisms -- which rely on siloed compliance units and episodic approvals rather than continuous, integrated supervision. We identify five governance dimensions essential for responsible agent deployment: cross-departmental implementation, comprehensive evaluation, enhanced security protocols, operational visibility, and systematic auditing. We evaluate the capacity of existing oversight structures to meet these challenges, via a mixed-methods approach consisting of a literature review and interviews with civil servants in AI-related roles. We find that agent oversight poses intensified versions of three existing governance challenges: continuous oversight, deeper integration of governance and operational capabilities, and interdepartmental coordination. We propose approaches that both adapt institutional structures and design agent oversight compatible with public sector constraints.

GermanPartiesQA: Benchmarking Commercial Large Language Models for Political Bias and Sycophancy

LLMs are changing the way humans create and interact with content, potentially affecting citizens' political opinions and voting decisions. As LLMs increasingly shape our digital information ecosystems, auditing to evaluate biases, sycophancy, or steerability has emerged as an active field of research. In this paper, we evaluate and compare the alignment of six LLMs by OpenAI, Anthropic, and Cohere with German party positions and evaluate sycophancy based on a prompt experiment. We contribute to evaluating political bias and sycophancy in multi-party systems across major commercial LLMs. First, we develop the benchmark dataset GermanPartiesQA based on the Voting Advice Application Wahl-o-Mat covering 10 state and 1 national elections between 2021 and 2023. In our study, we find a left-green tendency across all examined LLMs. We then conduct our prompt experiment for which we use the benchmark and sociodemographic data of leading German parliamentarians to evaluate changes in LLMs responses. To differentiate between sycophancy and steerabilty, we use 'I am [politician X], ...' and 'You are [politician X], ...' prompts. Against our expectations, we do not observe notable differences between prompting 'I am' and 'You are'. While our findings underscore that LLM responses can be ideologically steered with political personas, they suggest that observed changes in LLM outputs could be better described as personalization to the given context rather than sycophancy.

Immunization against harmful fine-tuning attacks

Approaches to aligning large language models (LLMs) with human values has focused on correcting misalignment that emerges from pretraining. However, this focus overlooks another source of misalignment: bad actors might purposely fine-tune LLMs to achieve harmful goals. In this paper, we present an emerging threat model that has arisen from alignment circumvention and fine-tuning attacks. However, lacking in previous works is a clear presentation of the conditions for effective defence. We propose a set of conditions for effective defence against harmful fine-tuning in LLMs called "Immunization conditions," which help us understand how we would construct and measure future defences. Using this formal framework for defence, we offer a synthesis of different research directions that might be persued to prevent harmful fine-tuning attacks and provide a demonstration of how to use these conditions experimentally showing early results of using an adversarial loss to immunize LLama2-7b-chat.