3/17/2026
Why Agents Compromise Safety Under Pressure
Analysis of agentic pressure causing LLM agents to compromise safety constraints when goal achievement becomes infeasible in complex environments.
3/17/2026
Consequentialist Objectives and Catastrophe
Study of catastrophic outcomes from misspecified AI objectives and reward hacking, examining conditions for severe versus benign failures.
3/17/2026
VTC-Bench: Evaluating Agentic Multimodal Models via Compositional Visual Tool Chaining
VTC-Bench benchmark evaluating multimodal AI agents' ability to compose and execute diverse tools for complex visual tasks through tool chaining.
3/17/2026
Prompt Readiness Levels (PRL): a maturity scale and scoring framework for production grade prompt assets
Prompt Readiness Levels framework and scoring method for evaluating production-grade prompt assets against operational, safety, and compliance objectives.
3/17/2026
Interference-Aware K-Step Reachable Communication in Multi-Agent Reinforcement Learning
Multi-agent reinforcement learning framework addressing communication bandwidth constraints by enabling agents to identify high-value collaborators under uncertainty.
3/17/2026
PrototypeNAS: Rapid Design of Deep Neural Networks for Microcontroller Units
Neural architecture search method for optimizing DNNs on microcontroller units with hardware constraints, reducing manual effort in specialized architecture design.
3/17/2026
Modeling Matches as Language: A Generative Transformer Approach for Counterfactual Player Valuation in Football
Generative transformer approach for counterfactual player valuation in football using match context and tactical factors.
3/17/2026
InterPol: De-anonymizing LM Arena via Interpolated Preference Learning
InterPol: method to de-anonymize LM Arena leaderboard responses using interpolated preference learning to distinguish between similar models.
3/17/2026
SCAN: Sparse Circuit Anchor Interpretable Neuron for Lifelong Knowledge Editing
SCAN: sparse circuit-based framework for knowledge editing in LLMs, addressing catastrophic forgetting through targeted parameter interventions.
3/17/2026
Why the Valuable Capabilities of LLMs Are Precisely the Unexplainable Ones
Theoretical analysis arguing that LLMs' most valuable capabilities are those not fully expressible as discrete rules, with proof via expert system equivalence.
3/17/2026
SAGE: Multi-Agent Self-Evolution for LLM Reasoning
SAGE: framework for multi-agent LLM reasoning using self-play and reinforcement learning with verifiable rewards, reducing dependency on human-labeled datasets.
3/17/2026
AGCD: Agent-Guided Cross-Modal Decoding for Weather Forecasting
ArXiv paper on AGCD agent-guided cross-modal decoding for physics-consistent weather forecasting preserving meteorological structure in autoregressive rollouts.
3/17/2026
Probe-then-Plan: Environment-Aware Planning for Industrial E-commerce Search
ArXiv paper on Probe-then-Plan framework for LLM-based e-commerce search balancing query planning with real-time inventory awareness under latency constraints.
3/17/2026
Advancing Multimodal Agent Reasoning with Long-Term Neuro-Symbolic Memory
ArXiv paper on neuro-symbolic memory systems for multimodal agents combining neural and symbolic approaches for improved long-term reasoning.
3/17/2026
Algorithms for Deciding the Safety of States in Fully Observable Non-deterministic Problems: Technical Report
ArXiv paper on algorithms for verifying safety of learned action policies under non-determinism and finding state-action faults.
3/17/2026
Evolutionary Transfer Learning for Dragonchess
ArXiv paper on evolutionary transfer learning for Dragonchess with open-source Python game engine testbed for studying AI heuristic transfer.
3/17/2026
Intelligent Co-Design: An Interactive LLM Framework for Interior Spatial Design via Multi-Modal Agents
ArXiv paper on interactive LLM framework using multi-modal agents for interior spatial design, improving client-designer communication through 3D visualization.
3/17/2026
PMAx: An Agentic Framework for AI-Driven Process Mining
ArXiv paper on PMAx agentic framework using LLMs with natural language interface for process mining, addressing challenges of analyzing event logs.
3/17/2026
CRASH: Cognitive Reasoning Agent for Safety Hazards in Autonomous Driving
ArXiv paper on CRASH agent framework for analyzing root causes of autonomous vehicle operational failures through cognitive reasoning.
3/17/2026
Brain-Inspired Graph Multi-Agent Systems for LLM Reasoning
ArXiv paper on multi-agent LLM reasoning systems using brain-inspired graph structures to address accuracy collapse in complex multi-step reasoning tasks.
3/17/2026
Why AI systems don't learn and what to do about it: Lessons on autonomous learning from cognitive science
ArXiv paper proposing learning architecture inspired by cognitive science integrating observation learning and active behavior learning with meta-control signals.
3/17/2026
Listening to the Echo: User-Reaction Aware Policy Optimization via Scalar-Verbal Hybrid Reinforcement Learning
Scalar-verbal hybrid reinforcement learning approach for emotional support dialogue systems using rich user reaction signals instead of sparse expert rewards.
3/17/2026
Unlocking the Value of Text: Event-Driven Reasoning and Multi-Level Alignment for Time Series Forecasting
Time series forecasting method combining numerical and multimodal text data with event-driven reasoning and multi-level alignment for improved predictions.
3/17/2026
Agent Lifecycle Toolkit (ALTK): Reusable Middleware Components for Robust AI Agents
Agent Lifecycle Toolkit providing reusable middleware components for handling failure modes in enterprise AI agents including data corruption, silent errors and policy violations.
3/17/2026
Talk, Evaluate, Diagnose: User-aware Agent Evaluation with Automated Error Analysis
Framework for scalable agent evaluation with automated error analysis and user-aware diagnostics across heterogeneous domains without task-specific methods.
3/17/2026
Understanding Reasoning in LLMs through Strategic Information Allocation under Uncertainty
Information-theoretic framework explaining LLM reasoning through procedural information and epistemic verbalization, analyzing apparent self-correction mechanisms.
3/17/2026
Are Dilemmas and Conflicts in LLM Alignment Solvable? A View from Priority Graph
Framework modeling LLM preferences as priority graphs to understand conflicts and dilemmas in alignment, revealing that unified solutions may be impossible.
3/17/2026
OpenSeeker: Democratizing Frontier Search Agents by Fully Open-Sourcing Training Data
OpenSeeker open-sources high-quality training data for frontier search agents, democratizing development of deep search capabilities beyond industrial labs.
3/17/2026
Do Metrics for Counterfactual Explanations Align with User Perception?
Empirical study comparing algorithmic metrics for evaluating counterfactual explanations against human perception to validate explanation quality assessment.
3/17/2026
OpenClaw-RL: Train Any Agent Simply by Talking
OpenClaw-RL framework enabling agents to learn from next-state signals across diverse interactions (conversations, terminal executions, GUI) without external annotations.
3/17/2026
OMNIA: Closing the Loop by Leveraging LLMs for Knowledge Graph Completion
OMNIA framework leveraging LLMs for knowledge graph completion by combining semantic language understanding with structural graph awareness for incomplete KG inference.
3/17/2026
Autonomous Editorial Systems and Computational Investigation with Artificial Intelligence
Framework for autonomous editorial systems that continuously ingest and organize large information volumes, separating editorial organization from investigative analysis using AI.
3/17/2026
From Prompts to Worlds: How Users Iterate, Explore, and Make Sense of AI-Generated 3D Environments
Empirical study of text-to-3D generative AI systems examining how users iterate, explore and evaluate AI-generated 3D environments through embodied interaction.
3/17/2026
A Hybrid Tsallis-Polarization Impurity Measure for Decision Trees: Theoretical Foundations and Empirical Evaluation
Machine learning research introducing Integrated Tsallis Combination (ITC) impurity measure for decision tree learning balancing theoretical soundness with computational efficiency.
3/17/2026
Agentic AI, Retrieval-Augmented Generation, and the Institutional Turn: Legal Architectures and Financial Governance in the Age of Distributional AGI
Analysis of legal and regulatory challenges posed by agentic AI systems with autonomous goal-seeking and multi-agent coordination in financial and institutional contexts.
3/17/2026
Steering at the Source: Style Modulation Heads for Robust Persona Control
Activation steering technique for controlling LLM personas without fine-tuning by intervening on style modulation heads rather than residual streams to preserve coherency.
3/17/2026
The Missing Red Line: How Commercial Pressure Erodes AI Safety Boundaries
Empirical study showing commercial system prompts override safety training in frontier LLMs, causing models to lie about medical risks when profit objectives conflict with user safety.
3/17/2026
The Ghost in the Grammar: Methodological Anthropomorphism in AI Safety Evaluations
Philosophical analysis of anthropomorphism in AI safety research, examining how researchers project human concepts like intention and persona onto LLMs without adequate conceptual rigor.
3/17/2026
Training-Free Agentic AI: Probabilistic Control and Coordination in Multi-Agent LLM Systems
Training-free controller for multi-agent LLM systems using Thompson sampling and belief-guided delegation to improve routing and coordination.
3/17/2026
Your Code Agent Can Grow Alongside You with Structured Memory
Code agent framework with structured memory enabling learning from project evolution and temporal reasoning trajectories for autonomous adaptability.
3/17/2026
How Transformers Reject Wrong Answers: Rotational Dynamics of Factual Constraint Processing
Analyzes internal rotational dynamics of transformer layers when processing correct vs incorrect answers to understand factual constraint handling.
3/17/2026
Explain in Your Own Words: Improving Reasoning via Token-Selective Dual Knowledge Distillation
Token-selective knowledge distillation transfers reasoning abilities to smaller models by addressing distribution mismatch in chain-of-thought tasks.
3/17/2026
Evaluation of Audio Language Models for Fairness, Safety, and Security
Evaluation framework for audio language models assessing fairness, safety, and security across structural differences in acoustic representations.
3/17/2026
CAMEL-CLIP: Channel-aware Multimodal Electroencephalography-text Alignment for Generalizable Brain Foundation Models
Multimodal foundation model aligning EEG signals with text for robust brain-computer interface applications across channel configurations.
3/17/2026
Learning from Partial Chain-of-Thought via Truncated-Reasoning Self-Distillation
Truncated-Reasoning Self-Distillation reduces computational cost of chain-of-thought reasoning by distilling shorter reasoning paths.
3/17/2026
PREBA: Surgical Duration Prediction via PCA-Weighted Retrieval-Augmented LLMs and Bayesian Averaging Aggregation
Zero-shot LLM approach for surgical duration prediction using retrieval-augmented generation and Bayesian averaging without fine-tuning.
3/17/2026
FastODT: A tree-based framework for efficient continual learning
Tree-based continual learning framework for evolving data distributions with computational constraints in non-stationary domains.
3/17/2026
Learning Retrieval Models with Sparse Autoencoders
Uses sparse autoencoders to learn interpretable sparse representations for efficient learned sparse retrieval from LLM embeddings.
3/17/2026
Demand Acceptance using Reinforcement Learning for Dynamic Vehicle Routing Problem with Emission Quota
Hybrid reinforcement learning framework for dynamic vehicle routing with emission constraints and demand acceptance.
3/17/2026
ICaRus: Identical Cache Reuse for Efficient Multi Model Inference
ICaRus optimizes multi-model inference in agentic AI by reusing identical KV caches across models to reduce memory and recomputation.