The GEST-Engine is a system that generates fully-annotated synthetic multi-actor video from natural language input by maintaining an explicit, inspectable world model represented as a Graph of Events in Space and Time (GEST). It produces frame-aligned RGB video, depth, segmentation, pose, and other annotations at zero marginal annotation cost. The system guarantees object permanence and temporal consistency, making it suitable for generating training data and evaluation benchmarks for video understanding.
Why it matters: GEST-Engine enables scalable production of richly annotated synthetic video with guaranteed consistency, potentially reducing reliance on manual annotation in video research.
A new method uses vision-language models (VLMs) to detect anomalous frames between maintenance task videos, enabling automatic extraction of expert-specific actions and contextual decision-making scenes. In simulated maintenance experiments, the approach achieved extraction rates of 65% for actions and 61% for decision-making scenes, outperforming conventional methods. The technique leverages frame-wise visual descriptions and intra-video self-similarity to identify key moments of expert know-how.
Why it matters: This method could facilitate the transfer of expert knowledge to less experienced workers by automatically identifying and extracting critical scenes from maintenance videos.
A new preprint examines the use of large language models (LLMs) for content moderation through 'policy-as-prompt' methods, where moderation policies are given to LLMs as natural-language prompts. The authors argue that this approach introduces specific risks and harms, and that simply writing prompts is not sufficient for effective or meaningful community governance. They propose several considerations for improving prompt governance but conclude that prompt-writing alone cannot ensure robust moderation outcomes.
Why it matters: This research highlights important limitations and governance challenges for AI-driven, prompt-based content moderation systems as their use expands in online communities.
A new preprint compares institutional and course-level generative AI policies at U.S. research-intensive universities. The study finds that while institutions tend to be more supportive of GenAI use, course-level guidance in computing education remains cautious. The authors propose an instructor-centered framework to guide future GenAI adoption in courses.
Why it matters: This research highlights a disconnect between university-wide AI policies and classroom practices, offering a framework to help computing educators navigate GenAI adoption.
A new preprint introduces DeGuNet, an ultra-compact image backbone designed for LiDAR-camera 3D detection in autonomous driving. DeGuNet uses depth-guided representation learning to address parameter redundancy in current multi-modal frameworks, reducing GPU memory usage by up to 66.5% and achieving a 1.16x speedup. The method also improves mean average precision (mAP) by up to 6.20 points on the nuScenes dataset, demonstrating both efficiency and accuracy gains.
Why it matters: DeGuNet could enable more efficient and accurate 3D perception for autonomous vehicles by significantly reducing computational demands.
A new preprint systematically evaluates continual learning (CL) methods for medical visual question answering (MedVQA) across a range of clinical tasks, including classification, detection, cell counting, and report generation. The study finds that current CL methods have difficulty maintaining a balance between retaining old knowledge and learning new tasks when faced with diverse objectives and supervision formats. The authors also analyze the impact of task ordering and the evolution of model parameters during continual learning. Code and experimental setup will be made publicly available.
Why it matters: This work exposes key limitations in current continual learning approaches for medical VQA, highlighting challenges that must be addressed for robust real-world deployment.
A new preprint introduces SymbOmni, an agentic omni-model designed to address the 'perpetual novice' problem in visual generation by leveraging Symbolic Concept Learning. The model features a Symbolic Concept Box that abstracts experiences into reusable instructions, enabling cumulative learning and compositional generalization. Experimental results show that SymbOmni outperforms existing agent-based and closed-source systems in image quality and task success, reduces token consumption by over 40%, and achieves state-of-the-art continual learning performance.
Why it matters: This work presents a novel approach for enabling AI models to learn cumulatively and evolve autonomously, potentially overcoming a key limitation of current monolithic models.
AutoTrace is an agentic pipeline that uses LLM agents to explore code property graphs and localize vulnerability triggers, even when they are located outside patched functions. On the InterPVD benchmark, AutoTrace achieves 75.0% VulnHit and 80.8% FuncHit, surpassing previous methods. The authors also introduce SinkTrace-Bench, a dataset of 1,542 source-to-sink causal chains, which reveals that current frontier LLMs struggle with causal reasoning in vulnerability analysis.
Why it matters: This work advances automated vulnerability analysis by enabling interprocedural trigger localization and provides a new benchmark that highlights the causal reasoning limitations of current LLMs.
A new preprint introduces StableAML, a machine learning framework for detecting money laundering in stablecoin transactions on Ethereum. The study finds that domain-informed tree ensemble models outperform graph neural networks in identifying suspicious wallets, and can distinguish between behavioral patterns of cybercrime syndicates and sanctioned entities. The approach is designed to support compliance with emerging regulations such as the EU's MiCA and the U.S. GENIUS Act.
Why it matters: This work presents a novel, interpretable method for high-precision behavioral classification in stablecoin anti-money laundering, potentially improving compliance and reducing unjustified asset freezes under new regulatory frameworks.
A new preprint introduces the agent:// URI scheme, designed to decouple agent identity from network location in multi-agent systems. The scheme incorporates trust roots, hierarchical capability paths, and sortable unique identifiers, with cryptographic attestation using PASETO tokens. Evaluation demonstrates full capability expressiveness on 369 tools, perfect discovery precision across 10,000 agents, and sub-5-microsecond performance, suggesting a robust and scalable approach to agent identity and discovery.
Why it matters: This work proposes a novel, decentralized solution to agent identity and discovery, potentially enabling more robust and scalable multi-agent ecosystems.
A preprint study evaluated 13 large language models (LLMs) on their ability to detect malicious Python packages from PyPI. The models achieved high recall and F1 scores up to 0.99 for flagging malicious packages, but struggled to accurately identify specific lines of malicious code, with weighted F1 dropping to 0.48. Code complexity, particularly longer packages, was found to be the main factor reducing performance.
Why it matters: The findings suggest LLMs can assist in initial triage for software supply chain security, but are not yet reliable for pinpointing specific malicious behaviors.
A new preprint introduces the One-Word Census, a minimal and low-cost method to measure answer-choice conformity across 44 language models. The study finds that when prompted to 'pick a word,' 41% of models chose 'serendipity,' and in several categories, a single answer dominated responses. The research reveals that newer flagship models tend to be the most conformist, while persona- and community-tuned models are more divergent. The findings highlight structured variation in conformity across model families and generations.
Why it matters: This work provides a practical tool for benchmarking diversity and conformity in language model outputs, with implications for understanding creativity and bias in AI systems.
A recent audit of inference-time defenses for multimodal large language models (MLLMs) found that three benchmark branches failed provenance checks, restricting reliable comparative results to only two datasets. The study also discovered that a legacy keyword-based safety protocol incorrectly counted empty strings as safe, with raw model responses unavailable to reassess this effect. Contrary to earlier claims, the archive does not support widespread model refusal, with a pooled refusal rate of just 0.52%.
Why it matters: This work exposes critical methodological issues in evaluating MLLM safety defenses, emphasizing the importance of traceable audits and provenance standards for trustworthy comparisons.
KnowAct-GUIClaw is a new framework for GUI automation that addresses limitations in cross-platform support and self-evolution found in previous systems like OpenClaw. The framework introduces a Know-Route-Act-Reflect paradigm, enabling the agent to accumulate user interaction experience and improve execution accuracy over time. In experiments, KnowAct-GUIClaw using the open-source Kimi-2.6 model achieved 64.1% on the MobileWorld benchmark, outperforming both open and closed-source alternatives such as Seed-2.0-Pro and GPT-5.5.
Why it matters: This work represents a notable advance in personal GUI assistants by enabling self-evolving memory and skill transferability across platforms and models, potentially improving automation efficiency and adaptability.
Researchers introduce QDEvo, a framework that integrates Quality-Diversity optimization with large language model (LLM)-driven heuristic search to address mode collapse in automated heuristic design. QDEvo maintains an archive of semantically diverse algorithms using code embeddings and hierarchical self-reflection. Experiments across benchmarks and real-world applications show that QDEvo outperforms state-of-the-art methods in Hypervolume and Inverted Generational Distance metrics.
Why it matters: This work advances automated heuristic design by promoting semantic diversity, potentially enabling more robust and efficient solutions for complex optimization problems.
A new arXiv preprint introduces an information-theoretic framework for watermark forensics in generative models, establishing fundamental limits on detection, attribution, and payload extraction. The authors derive tight entropy-rate laws, showing that attributing text to one of N users requires Θ(log N/h) tokens and extracting an ℓ-bit payload requires Θ(ℓ/h) tokens, where h is the entropy rate. Experiments on GPT-2, Pythia-410M, and Qwen2.5 confirm the theoretical predictions.
Why it matters: This work rigorously quantifies the trade-offs and costs of watermarking in generative AI, providing a theoretical foundation that can inform the design of more reliable forensic tools.
A new study introduces a Bayesian model for intercomprehension—the partial understanding of an unfamiliar but related language—by leveraging a first-language (L1) language model and a noise model to infer mappings from the unfamiliar language (L2) to L1. Human experiments demonstrate that the model's predictions closely match human performance and outperform zero-shot prompting of much larger language models.
Why it matters: This work advances our understanding of cross-language comprehension and provides a computational framework that could inform the development of more flexible multilingual AI systems.
Researchers present MCP Pitfall Lab, a protocol-aware security testing framework for Model Context Protocol (MCP) tool servers that models developer pitfalls as reproducible scenarios and validates outcomes using objective validators. In 2,579 runs across four models, they report a 31.9% overall attack success rate, with multi-modal injection attacks achieving the highest rate at 38.7%. The framework also introduces the Semantic MCP-Bill-of-Material (MCP-BOM), which augments component inventories with security-relevant tool semantics to aid in auditing and hardening. The study demonstrates that static analysis alone is insufficient for certain attack vectors, highlighting the need for runtime provenance.
Why it matters: This work provides a novel and practical framework for systematically identifying and mitigating security risks in the MCP tool server ecosystem, advancing the security of AI software supply chains.
Researchers have introduced a practical evaluation protocol for AI pentesting agents that emphasizes validated vulnerability discovery in complex, real-world targets, rather than just task completion. The protocol incorporates LLM-based semantic matching, bipartite resolution, and cumulative scoring to enable more operationally meaningful comparisons between agents. Expert-annotated ground truth and code are released to support reproducibility.
Why it matters: This protocol bridges the gap between controlled benchmarks and real-world offensive security, supporting more reliable and realistic assessment of AI agents for cybersecurity applications.
A new preprint introduces the Roundtable Context Window Test (RCWT), a protocol designed to measure how coordination content—such as shared state and prior discussion—displaces task-relevant tokens in large language model (LLM) prompts under fixed context budgets. The study finds that as coordination overhead increases, model performance drops sharply once the remaining task-relevant evidence falls below a few hundred tokens. An ablation experiment shows this effect is due to budget displacement, not semantic interference from coordination content.
Why it matters: This work provides a practical measurement tool for understanding and managing context allocation in multi-agent and memory-augmented LLM systems, highlighting a key bottleneck when coordination content consumes limited prompt space.