Existing evaluations of AI misuse safeguards provide a patchwork of evidence that is often difficult to connect to real-world decisions. To bridge this gap, we describe an end-to-end argument (a "safety case") that misuse safeguards reduce the risk posed by an AI assistant to low levels. We first describe how a hypothetical developer red teams safeguards, estimating the effort required to evade them. Then, the developer plugs this estimate into a quantitative "uplift model" to determine how much barriers introduced by safeguards dissuade misuse (https://www.aimisusemodel.com/). This procedure provides a continuous signal of risk during deployment that helps the developer rapidly respond to emerging threats. Finally, we describe how to tie these components together into a simple safety case. Our work provides one concrete path -- though not the only path -- to rigorously justifying AI misuse risks are low.

Dataset transparency is a key enabler of responsible AI, but insights into multimodal dataset attributes that impact trustworthy and ethical aspects of AI applications remain scarce and are difficult to compare across datasets. To address this challenge, we introduce Trustworthy and Ethical Dataset Indicators (TEDI) that facilitate the systematic, empirical analysis of dataset documentation. TEDI encompasses 143 fine-grained indicators that characterize trustworthy and ethical attributes of multimodal datasets and their collection processes. The indicators are framed to extract verifiable information from dataset documentation. Using TEDI, we manually annotated and analyzed over 100 multimodal datasets that include human voices. We further annotated data sourcing, size, and modality details to gain insights into the factors that shape trustworthy and ethical dimensions across datasets. We find that only a select few datasets have documented attributes and practices pertaining to consent, privacy, and harmful content indicators. The extent to which these and other ethical indicators are addressed varies based on the data collection method, with documentation of datasets collected via crowdsourced and direct collection approaches being more likely to mention them. Scraping dominates scale at the cost of ethical indicators, but is not the only viable collection method. Our approach and empirical insights contribute to increasing dataset transparency along trustworthy and ethical dimensions and pave the way for automating the tedious task of extracting information from dataset documentation in future.

As foundation models grow increasingly more intelligent, reliable and trustworthy safety evaluation becomes more indispensable than ever. However, an important question arises: Whether and how an advanced AI system would perceive the situation of being evaluated, and lead to the broken integrity of the evaluation process? During standard safety tests on a mainstream large reasoning model, we unexpectedly observe that the model without any contextual cues would occasionally recognize it is being evaluated and hence behave more safety-aligned. This motivates us to conduct a systematic study on the phenomenon of evaluation faking, i.e., an AI system autonomously alters its behavior upon recognizing the presence of an evaluation context and thereby influencing the evaluation results. Through extensive experiments on a diverse set of foundation models with mainstream safety benchmarks, we reach the main finding termed the observer effects for AI: When the AI system under evaluation is more advanced in reasoning and situational awareness, the evaluation faking behavior becomes more ubiquitous, which reflects in the following aspects: 1) Reasoning models recognize evaluation 16% more often than non-reasoning models. 2) Scaling foundation models (32B to 671B) increases faking by over 30% in some cases, while smaller models show negligible faking. 3) AI with basic memory is 2.3x more likely to recognize evaluation and scores 19% higher on safety tests (vs. no memory). To measure this, we devised a chain-of-thought monitoring technique to detect faking intent and uncover internal signals correlated with such behavior, offering insights for future mitigation studies.

Temporal Knowledge Graphs (TKGs) represent dynamic facts as timestamped relations between entities. TKG completion involves forecasting missing or future links, requiring models to reason over time-evolving structure. While LLMs show promise for this task, existing approaches often overemphasize supervised fine-tuning and struggle particularly when historical evidence is limited or missing. We introduce RECIPE-TKG, a lightweight and data-efficient framework designed to improve accuracy and generalization in settings with sparse historical context. It combines (1) rule-based multi-hop retrieval for structurally diverse history, (2) contrastive fine-tuning of lightweight adapters to encode relational semantics, and (3) test-time semantic filtering to iteratively refine generations based on embedding similarity. Experiments on four TKG benchmarks show that RECIPE-TKG outperforms previous LLM-based approaches, achieving up to 30.6\% relative improvement in Hits@10. Moreover, our proposed framework produces more semantically coherent predictions, even for the samples with limited historical context.

Jailbreak attacks have been observed to largely fail against recent reasoning models enhanced by Chain-of-Thought (CoT) reasoning. However, the underlying mechanism remains underexplored, and relying solely on reasoning capacity may raise security concerns. In this paper, we try to answer the question: Does CoT reasoning really reduce harmfulness from jailbreaking? Through rigorous theoretical analysis, we demonstrate that CoT reasoning has dual effects on jailbreaking harmfulness. Based on the theoretical insights, we propose a novel jailbreak method, FicDetail, whose practical performance validates our theoretical findings.

--This paper presents an adaptive framework for edge inference based on a dynamically configurable transformer-powered deep joint source channel coding (DJSCC) architecture. Motivated by a practical scenario where a resource constrained edge device engages in goal oriented semantic communication, such as selectively transmitting essential features for object detection to an edge server, our approach enables efficient task aware data transmission under varying bandwidth and channel conditions. T o achieve this, input data is tokenized into compact high level semantic representations, refined by a transformer, and transmitted over noisy wireless channels. As part of the DJSCC pipeline, we employ a semantic token selection mechanism that adaptively compresses informative features into a user specified number of tokens per sample. These tokens are then further compressed through the JSCC module, enabling a flexible token communication strategy that adjusts both the number of transmitted tokens and their embedding dimensions. We incorporate a resource allocation algorithm based on Lyapunov stochastic optimization to enhance robustness under dynamic network conditions, effectively balancing compression efficiency and task performance. Experimental results demonstrate that our system consistently outperforms existing baselines, highlighting its potential as a strong foundation for AI native semantic communication in edge intelligence applications. ECENTL Y, there was a surge of interest in semantic and goal-oriented communications, establishing this paradigm as crucial for the development of 6G AI-native communication networks [1], [2]. A. Devoto is with the Department of Computer, Control, and Management Engineering (DIAG) at Sapienza University of Rome, Italy. He is also a member of the Consorzio Nazionale Interuniversitario per le Telecomunicazioni (CNIT), Parma, Italy. Jary Pomponi, Paolo Di Lorenzo, and Simone Scardapane are with the Department of Information Engineering, Electronics, and Telecommunications (DIET) at Sapienza University of Rome, Italy, and they are also affiliated with CNIT. Mat-tia Merluzzi is with CEA-Leti, Universit e Grenoble Alpes, located in Grenoble, France. This work has been supported by the SNS JU project 6G-GOALS under the EU's Horizon program Grant Agreement No 101139232, by Sapienza grant RG123188B3EF6A80 (CENTS), and by European Union under the Italian National Recovery and Resilience Plan of NextGenerationEU, partnership on Telecommunications of the Future (PE00000001 - program REST ART).

General-purpose large language models, despite their broad capabilities, often struggle with specialized domain knowledge, a limitation particularly pronounced in more accessible, lower-parameter versions. This gap hinders their deployment as effective agents in demanding fields such as astronomy. Building on our prior work with AstroSage-8B, this study introduces AstroSage-70B, a significantly larger and more advanced domain-specialized natural-language AI assistant. It is designed for research and education across astronomy, astrophysics, space science, astroparticle physics, cosmology, and astronomical instrumentation. Developed from the Llama-3.1-70B foundation, AstroSage-70B underwent extensive continued pre-training on a vast corpus of astronomical literature, followed by supervised fine-tuning and model merging. Beyond its 70-billion parameter scale, this model incorporates refined datasets, judiciously chosen learning hyperparameters, and improved training procedures, achieving state-of-the-art performance on complex astronomical tasks. Notably, we integrated reasoning chains into the SFT dataset, enabling AstroSage-70B to either answer the user query immediately, or first emit a human-readable thought process. Evaluated on the AstroMLab-1 benchmark -- comprising 4,425 questions from literature withheld during training -- AstroSage-70B achieves state-of-the-art performance. It surpasses all other tested open-weight and proprietary models, including leading systems like o3, Gemini-2.5-Pro, Claude-3.7-Sonnet, Deepseek-R1, and Qwen-3-235B, even those with API costs two orders of magnitude higher. This work demonstrates that domain specialization, when applied to large-scale models, can enable them to outperform generalist counterparts in specialized knowledge areas like astronomy, thereby advancing the frontier of AI capabilities in the field.

Algorithmic decision-making systems sometimes produce errors or skewed predictions toward a particular group, leading to unfair results. Debiasing practices, applied at different stages of the development of such systems, occasionally introduce new forms of unfairness or exacerbate existing inequalities. We focus on post-processing techniques that modify algorithmic predictions to achieve fairness in classification tasks, examining the unintended consequences of these interventions. To address this challenge, we develop a set of measures that quantify the disparity in the flips applied to the solution in the post-processing stage. The proposed measures will help practitioners: (1) assess the proportionality of the debiasing strategy used, (2) have transparency to explain the effects of the strategy in each group, and (3) based on those results, analyze the possibility of the use of some other approaches for bias mitigation or to solve the problem. We introduce a methodology for applying the proposed metrics during the post-processing stage and illustrate its practical application through an example. This example demonstrates how analyzing the proportionality of the debiasing strategy complements traditional fairness metrics, providing a deeper perspective to ensure fairer outcomes across all groups.

LLM-based digital twin simulation, where large language models are used to emulate individual human behavior, holds great promise for research in AI, social science, and digital experimentation. However, progress in this area has been hindered by the scarcity of real, individual-level datasets that are both large and publicly available. This lack of high-quality ground truth limits both the development and validation of digital twin methodologies. To address this gap, we introduce a large-scale, public dataset designed to capture a rich and holistic view of individual human behavior. We survey a representative sample of $N = 2,058$ participants (average 2.42 hours per person) in the US across four waves with 500 questions in total, covering a comprehensive battery of demographic, psychological, economic, personality, and cognitive measures, as well as replications of behavioral economics experiments and a pricing survey. The final wave repeats tasks from earlier waves to establish a test-retest accuracy baseline. Initial analyses suggest the data are of high quality and show promise for constructing digital twins that predict human behavior well at the individual and aggregate levels. By making the full dataset publicly available, we aim to establish a valuable testbed for the development and benchmarking of LLM-based persona simulations. Beyond LLM applications, due to its unique breadth and scale the dataset also enables broad social science research, including studies of cross-construct correlations and heterogeneous treatment effects.

Large language models (LLMs) have demonstrated impressive capabilities in reasoning with the emergence of reasoning models like OpenAI-o1 and DeepSeek-R1. Recent research focuses on integrating reasoning capabilities into the realm of retrieval-augmented generation (RAG) via outcome-supervised reinforcement learning (RL) approaches, while the correctness of intermediate think-and-search steps is usually neglected. To address this issue, we design a process-level reward module to mitigate the unawareness of intermediate reasoning steps in outcome-level supervision without additional annotation. Grounded on this, we propose Learning to Think-and-Search (LeTS), a novel framework that hybridizes stepwise process reward and outcome-based reward to current RL methods for RAG. Extensive experiments demonstrate the generalization and inference efficiency of LeTS across various RAG benchmarks. In addition, these results reveal the potential of process- and outcome-level reward hybridization in boosting LLMs' reasoning ability via RL under other scenarios. The code will be released soon.