STRIATUM-CTF is a modular agentic framework that uses the Model Context Protocol to mediate between large language model reasoning and the execution of offensive‑security tools. By exposing tools as strictly typed, schema‑validated functions, the protocol rejects invalid tool calls and returns structured observations that the agent can use to self‑correct. In controlled experiments the system achieved 77.8–86.7% success across 15 CTF challenges, and in a live multi‑team CTF at Florida State University it placed first among more than 30 teams.
By James Hugglestone, Samuel Jacob Chacko, Dawson Stoller, Ryan Schmidt, Xiuwen Liu
Long-range 3D object detection remains difficult for autonomous vehicles because LiDAR returns become sparse and camera evidence weakens with distance, while automotive radar keeps returning measureme...
Conventional recommender systems operate within platform-specific catalogs, limiting access to potentially valuable resources across platforms. We propose user-centered agentic recommendation, a parad...
Harnesses for large language models have converged on a single-threaded shape: one agent, one history, one machine, one sitting. That shape fails work that spans many agents, outlives any one model ca...
The paper demonstrates that a model‑based reinforcement‑learning agent can learn the causal structure of Minkowski spacetime solely from maximizing quantum mutual information between localized probes, without explicit knowledge of the metric. During training, the agent builds an internal predictive model whose planner develops the null structure characteristic of spacetime, enabling accurate reconstruction of the causal structure. This result suggests that spacetime concepts can emerge from an agent’s operational quantum experience.
HIVE-Diff is an event-conditioned video diffusion framework that reconstructs high‑frame‑rate HDR video from a single low‑dynamic‑range (LDR) reference frame and event data. It separates the task into event‑guided temporal generation and HDR content completion, enabling the creation of multiple HDR frames from one LDR observation. The model is trained on synthetic data and adapted to real events using teacher‑student distillation with LDR‑event sequences as proxy supervision, achieving strong reconstruction quality across synthetic and real datasets.
The paper examines how a memory‑augmented agent can determine the value of its memories using only retrieval logs. It shows that top‑k retrieval fails to assign any value to memories regardless of the number of episodes logged, and that a single question can determine at most one memory’s value. To address this, the authors propose LEDGER, which estimates determined Shapley values with confidence intervals, merges memories with identical histories, and strategically retrieves to increase determinability, achieving a substantial rise in determined memories on HotpotQA and LoCoMo.
The paper investigates whether coding agents can continually evolve across multiple software engineering tasks without catastrophic forgetting. By testing two harness evolution methods on six evolution directions derived from three datasets, the study finds that evolving on new datasets can drastically degrade performance on previously learned ones. The authors provide insights and directions for future continual self‑evolution of agent harnesses.
CUA-Master introduces a nested double-loop recursive improvement framework to scale computer-use agents. The outer flywheel continuously integrates open-world data and model weaknesses to guide data construction and optimization, while the inner flywheel mines instructional videos for seed tasks and propagates them into reusable primitives for diverse task synthesis. The system also employs a feedback-guided snapshot-rollback retry mechanism for reliable trajectory collection, achieving competitive performance on multiple benchmarks and demonstrating continuous gains across iterations.
Simon Willison discusses how the U.S. Census Bureau’s Business Trends and Outlook Survey has changed its AI‑adoption question wording, leading to a marked rise in reported AI usage among large companies. Using updated October 2026 data, he shows that firms with 250+ employees now report AI use in any business function at about 40%, up from roughly 12% in September 2025 when the question focused on producing goods or services. He notes that this shift coincides with the Bureau’s compliance with Executive Order 14434, the so‑called Super Intelligence order.
LUCA is a configurable synthetic world designed to evaluate experimental reasoning in agents by providing fully known ground truth. In this environment, agents learn unknown regulatory systems through budgeted experimentation, then predict unseen interventions, diagnose hidden faults, and attempt rescues, allowing separate assessment of experiment selection, prediction, diagnosis, and intervention. Across seven frontier LLMs and three baselines, the study shows that diagnosis performance varies widely while prediction remains relatively stable, and that agents often adopt different experimental strategies without a clear link between exploration and success.
Let Decision Models Watch Your Agents is a hands‑on guide that explains how to detect behavioral anomalies by analyzing an agent’s evolving execution history. The article provides practical steps and techniques for monitoring agent behavior over time. It is published on Towards Data Science.
By Shuai Guo
GraTED is a Graph–Tensor EEG Decoding Model that preserves the spatial and temporal structure of passenger electroencephalography (EEG) data by modeling electrode positions as an undirected graph and applying graph convolution and low‑rank tensor classification. It is designed to predict risk perception and identify danger in autonomous‑vehicle scenarios. On the PassengerEEG dataset, GraTED outperforms seven baseline methods in balanced accuracy while using the fewest learnable parameters.
RxnMiner is a multimodal agent that extracts structured reaction data from chemistry figures by combining expert-informed procedures with a chemistry toolbox. It reads molecular structures, resolves shared and entry-specific evidence, and reconstructs reaction records from templates. On the ChemRxn-Bench benchmark, RxnMiner improves stereochemistry-aware reaction macro‑F1 from 21.0% to 71.0%, outperforming existing systems and generating a large silver corpus of 88,202 reaction records from 10,107 figures.
AffToM is a new benchmark for affective Theory of Mind (ToM) in dynamic multi‑party interactions, comprising 1,922 test pairs and 19,759 training pairs. It evaluates Individual Affective Reasoning (IAR) and Multi‑party Affective Reasoning (MPAR), the latter tracking nested beliefs and emotions as events unfold. The authors introduce an event‑driven multi‑agent generation pipeline and a two‑stage post‑training framework, AffToM‑R1, which boosts MPAR accuracy by 13.38 percentage points over baseline models.
The paper introduces SYNTRA, a self‑improving data synthesis framework that generates reusable synthesis instructions from a few retrieval examples. By inferring implicit information needs and relevance criteria, SYNTRA refines query generation and relevance assessment through feedback, producing effective training data for embedding models across 24 datasets and 15 domains. Experiments show that SYNTRA improves retrieval performance, often surpassing human calibration, and its benefits scale with larger synthetic training sets, such as on MS MARCO.
By Zucan Lyu, Rui Tang, Jianlyu Chen, Peisheng Jin, Defu Lian
The paper proposes a portable specification framework for the infrastructure—called the harness—that surrounds large language models. It decomposes the harness into eight functional components, enabling consistent specification and comparison across systems. The authors argue that recording which model ran in which harness is essential because different configurations can drastically alter performance and expose varying attack surfaces, and current model or agent cards do not capture this sequence of effective configurations.
By Nicolas Gertler, Claudio Novelli, Luciano Floridi
The paper investigates how model-based reinforcement learning can support continual learning by reusing prior knowledge. It introduces a contextual Markov decision process that separates behavioral and physical context, and studies zero‑shot reuse in FourRooms tasks using Dyna‑Q and TD‑MPC. The experiments show that zero‑shot reuse degrades under physical context changes and that this degradation is strongly tied to the observation representation, highlighting representation generalization as a key bottleneck.
By Amir Mesbah, Di Wu
CausalXRL is a new framework that makes reinforcement learning agents explainable by using causal graph reasoning. It offers transparent, structured, multi‑level explanations of agent decisions. The authors validate the approach with quantitative tests on benchmark RL environments and a qualitative expert study on the real‑time strategy game MicroRTS, showing faithful explanations and efficient causal‑structure learning.
By Yanming Zhang, Eric Papenhausen, Klaus Mueller
The paper investigates how plasticity loss affects deep reinforcement learning in stochastic environments. By training SAC agents on DeepMind Control tasks with varying observation and action noise, the authors find that plasticity-preserving techniques improve robustness, especially as noise increases, while resets combined with layer normalization perform worse. They also observe that stochasticity can accelerate plasticity loss in harder tasks, but in easier settings noise can act as a regularizer, and that plasticity loss is most pronounced early in training, with TD instability dominating in the most challenging scenarios.
By Philipp Bordne, Andr\'e Biedenkapp