Agentic AI systems, software agents with autonomy, decision-making ability, and adaptability, are increasingly used to execute complex tasks on behalf of organisations. Most such systems rely on Large Language Models (LLMs), whose broad semantic capabilities enable powerful language processing but lack explicit, institution-specific grounding. In enterprises, data rarely comes with an inspectable semantic layer, and constructing one typically requires labour-intensive "data archaeology": cleaning, modelling, and curating knowledge into ontologies, taxonomies, and other formal structures. At the same time, explainability methods such as saliency maps expose an "interpretability gap": they highlight what the model attends to but not why, leaving decision processes opaque. In this preprint, we present a case study, developed by Kaiasm and Avantra AI through their work with The Turing Way Practitioners Hub, a forum developed under the InnovateUK BridgeAI program. This study presents a collaborative human-AI approach to building an inspectable semantic layer for Agentic AI. AI agents first propose candidate knowledge structures from diverse data sources; domain experts then validate, correct, and extend these structures, with their feedback used to improve subsequent models. Authors show how this process captures tacit institutional knowledge, improves response quality and efficiency, and mitigates institutional amnesia. We argue for a shift from post-hoc explanation to justifiable Agentic AI, where decisions are grounded in explicit, inspectable evidence and reasoning accessible to both experts and non-specialists.

The rapid progress of Large Language Models (LLMs) has transformed natural language processing and broadened its impact across research and society. Yet, systematic evaluation of these models, especially for languages beyond English, remains limited. "Challenging the Abilities of LAnguage Models in ITAlian" (CALAMITA) is a large-scale collaborative benchmarking initiative for Italian, coordinated under the Italian Association for Computational Linguistics. Unlike existing efforts that focus on leaderboards, CALAMITA foregrounds methodology: it federates more than 80 contributors from academia, industry, and the public sector to design, document, and evaluate a diverse collection of tasks, covering linguistic competence, commonsense reasoning, factual consistency, fairness, summarization, translation, and code generation. Through this process, we not only assembled a benchmark of over 20 tasks and almost 100 subtasks, but also established a centralized evaluation pipeline that supports heterogeneous datasets and metrics. We report results for four open-weight LLMs, highlighting systematic strengths and weaknesses across abilities, as well as challenges in task-specific evaluation. Beyond quantitative results, CALAMITA exposes methodological lessons: the necessity of fine-grained, task-representative metrics, the importance of harmonized pipelines, and the benefits and limitations of broad community engagement. CALAMITA is conceived as a rolling benchmark, enabling continuous integration of new tasks and models. This makes it both a resource -- the most comprehensive and diverse benchmark for Italian to date -- and a framework for sustainable, community-driven evaluation. We argue that this combination offers a blueprint for other languages and communities seeking inclusive and rigorous LLM evaluation practices.

AI policy guidance is predominantly written as prose, which practitioners must first convert into executable rules before frameworks can evaluate or enforce them. This manual step is slow, error-prone, difficult to scale, and often delays the use of safeguards in real-world deployments. To address this gap, we present Policy-to-Tests (P2T), a framework that converts natural-language policy documents into normalized, machine-readable rules. The framework comprises a pipeline and a compact domain-specific language (DSL) that encodes hazards, scope, conditions, exceptions, and required evidence, yielding a canonical representation of extracted rules. To test the framework beyond a single policy, we apply it across general frameworks, sector guidance, and enterprise standards, extracting obligation-bearing clauses and converting them into executable rules. These AI-generated rules closely match strong human baselines on span-level and rule-level metrics, with robust inter-annotator agreement on the gold set. To evaluate downstream behavioral and safety impact, we add HIPAA-derived safeguards to a generative agent and compare it with an otherwise identical agent without guardrails. An LLM-based judge, aligned with gold-standard criteria, measures violation rates and robustness to obfuscated and compositional prompts. Detailed results are provided in the appendix. We release the codebase, DSL, prompts, and rule sets as open-source resources to enable reproducible evaluation.

The emergence of open data portals necessitates more attention to protecting sensitive data before datasets get published and exchanged. While an abundance of methods for suppressing sensitive data exist, the conceptualization of sensitive data and methods to detect it, focus particularly on personal data that, if disclosed, may be harmful or violate privacy. We observe the need for refining and broadening our definitions of sensitive data, and argue that the sensitivity of data depends on its context. Based on this definition, we introduce two mechanisms for contextual sensitive data detection that consider the broader context of a dataset at hand. First, we introduce type contextualization, which first detects the semantic type of particular data values, then considers the overall context of the data values within the dataset or document. Second, we introduce domain contextualization which determines sensitivity of a given dataset in the broader context based on the retrieval of relevant rules from documents that specify data sensitivity (e.g., data topic and geographic origin). Experiments with these mechanisms, assisted by large language models (LLMs), confirm that: 1) type-contextualization significantly reduces the number of false positives for type-based sensitive data detection and reaches a recall of 94% compared to 63% with commercial tools, and 2) domain-contextualization leveraging sensitivity rule retrieval is effective for context-grounded sensitive data detection in non-standard data domains such as humanitarian datasets. Evaluation with humanitarian data experts also reveals that context-grounded LLM explanations provide useful guidance in manual data auditing processes, improving consistency. We open-source mechanisms and annotated datasets for contextual sensitive data detection at https://github.com/trl-lab/sensitive-data-detection.

Access to justice remains a global challenge, with many citizens still finding it difficult to seek help from the justice system when facing legal issues. Although the internet provides abundant legal information and services, navigating complex websites, understanding legal terminology, and filling out procedural forms continue to pose barriers to accessing justice. This paper introduces the LegalWebAgent framework that employs a web agent powered by multimodal large language models to bridge the gap in access to justice for ordinary citizens. The framework combines the natural language understanding capabilities of large language models with multimodal perception, enabling a complete process from user query to concrete action. It operates in three stages: the Ask Module understands user needs through natural language processing; the Browse Module autonomously navigates webpages, interacts with page elements (including forms and calendars), and extracts information from HTML structures and webpage screenshots; the Act Module synthesizes information for users or performs direct actions like form completion and schedule booking. To evaluate its effectiveness, we designed a benchmark test covering 15 real-world tasks, simulating typical legal service processes relevant to Québec civil law users, from problem identification to procedural operations. Evaluation results show LegalWebAgent achieved a peak success rate of 86.7%, with an average of 84.4% across all tested models, demonstrating high autonomy in complex real-world scenarios.

AI agents can autonomously perform tasks and, often without explicit user consent, collect or disclose users' sensitive local data, which raises serious privacy concerns. Although AI agents' privacy policies describe their intended data practices, there remains limited transparency and accountability about whether runtime behavior matches those policies. To close this gap, we introduce AudAgent, a visual tool that continuously monitors AI agents' data practices in real time and guards compliance with stated privacy policies. AudAgent consists of four components for automated privacy auditing of AI agents. (i) Policy formalization: a novel cross-LLM voting mechanism to guarantee confidence of the parsed privacy policy model. (ii) Runtime annotation: a lightweight Presidio-based analyzer detects sensitive data and annotates data practices based on the AI agent's context and the privacy policy model. (iii) Compliance auditing: ontology graphs and automata-based checking connect the privacy policy model with runtime annotations, enabling on-the-fly compliance checking. (iv) User interface: an infrastructure-independent implementation visualizes the real-time execution trace of AI agents along with potential privacy policy violations, providing user-friendly transparency and accountability. We evaluate AudAgent with AI agents built using mainstream frameworks, demonstrating its effectiveness in detecting and visualizing privacy policy violations in real time. Using AudAgent, we also find that most privacy policies omit explicit safeguards for highly sensitive data such as SSNs, whose misuse violates legal requirements, and that many agents do not refuse handling such data via third-party tools, including those controlled by Claude, Gemini, and DeepSeek. AudAgent proactively blocks operations on such data, overriding the agents' original privacy policy and behavior.

Alignment of large language models (LLMs) with principles like helpfulness, honesty, and harmlessness typically relies on scalar rewards that obscure which objectives drive the training signal. We introduce QA-LIGN, which decomposes monolithic rewards into interpretable principle-specific evaluations through structured natural language programs. Models learn through a draft, critique, and revise pipeline, where symbolic evaluation against the rubrics provides transparent feedback for both initial and revised responses during GRPO training. Applied to uncensored Llama-3.1-8B-Instruct, QA-LIGN reduces attack success rates by up to 68.7% while maintaining a 0.67% false refusal rate, achieving Pareto optimal safety-helpfulness performance and outperforming both DPO and GRPO with state-of-the-art reward models given equivalent training. These results demonstrate that making reward signals interpretable and modular improves alignment effectiveness, suggesting transparency enhances LLM safety.

Detecting vulnerabilities in source code remains a critical yet challenging task, especially when benign and vulnerable functions share significant similarities. In this work, we introduce VulTrial, a courtroom-inspired multi-agent framework designed to identify vulnerable code and to provide explanations. It employs four role-specific agents, which are security researcher, code author, moderator, and review board. Using GPT-4o as the base LLM, VulTrial almost doubles the efficacy of prior best-performing baselines. Additionally, we show that role-specific instruction tuning with small quantities of data significantly further boosts VulTrial's efficacy. Our extensive experiments demonstrate the efficacy of VulTrial across different LLMs, including an open-source, in-house-deployable model (LLaMA-3.1-8B), as well as the high quality of its generated explanations and its ability to uncover multiple confirmed zero-day vulnerabilities in the wild.

By rolling back auto industry fuel efficiency goals, US president Donald Trump hopes to make new cars cheaper. But prices won't drop for years, and consumers will spend more on gas in the meantime. The Trump administration says its proposal to roll back vehicle fuel economy standards, announced officially in the Oval Office on Wednesday, is an attempt to shave dollars off the ballooning cost of new cars in the US. But the intended price drops likely won't show up on dealership lots and showroom floors for months if not years, given the length of automakers' product planning schedule. It would also likely force Americans to pay more, long-term, at another place they tend to visit more frequently: the pump.

The Industry Trump Gave the Tech Bros Everything. Why Are They Still Crashing Out? This was supposed to be their year--but a historically unpopular president and fears of an A.I. stock market crash loom large over Silicon Valley. Enter your email to receive alerts for this author. You can manage your newsletter subscriptions at any time.