Recent research exposes critical fragilities in AI safety and reasoning, where auditors detect sabotage with only 0.59 AUROC and persona controls drop monitorability from 0.94 to 0.22. Overthinking reduces reasoning accuracy by 3.6 points, while nuclear engineering LLMs show weak reasoning despite high accuracy. DiscoveryAgent identifies 14x more errors than manual review, and MedGuardBench proves personalized guardrails significantly outperform standard ones in medication safety. Formal counterfactual explanations offer efficient computation guarantees, and WebAgents block 716 injected contract violations while maintaining portability.
Advances in agent infrastructure and memory management drive substantial efficiency gains, with SLED achieving 2,300x latency improvement over conventional slicing and Readie reducing sandbox startup latency by 3–22×. RegimeCouncil improves accuracy by 5.2–7.3 points by resolving memory conflicts, and EvoDisco gains up to +17.9 points via evolutionary operators. MathEye solves 93.3% of benchmarks with 55% fewer interactions, while ATLAS-Finance reveals frontier agents pass fewer than 11% of realistic banking tasks. Context management in coding agents increases costs by 33–42% despite reducing commands, highlighting optimization trade-offs.
Domain-specific applications demonstrate robust performance in specialized tasks, including a hybrid transport solver achieving a benefit of 487,237 and PeriaTrace diagnosing teeth with 94.37% F1. Abelio outperforms frontier LLMs in hypothesis diversity for epithelial repair, and VITAL offers a risk-adjusted framework for pharmaceutical trials. Autonomous driving models achieve 0.745m ADE, while IoT fuzzing uncovers smart home vulnerabilities. Climate downscaling eliminates conservation errors, and quantum chemistry calculations accelerate 77x via foresighted agentic computing.
Emerging techniques address alignment, verification, and cross-environment transfer, with DACE improving robustness against out-of-distribution attacks and ConfGuard blocking sensor attacks on predictive maintenance agents. IfcGym enables dependency-aware IFC editing with 86.78% success, and SYMAR accelerates verification by 108x. Selective maintenance prevents skill degradation by up to 67pp, and LMLineage detects concept drift to compact irrelevant context. Agent Trust Infrastructure increases attack containment from 43% to 84%, and EDA achieves 91.02% cue accuracy for depression analysis.
Key Takeaways
- Auditors detect AI sabotage with only 0.59 AUROC.
- Persona controls drop monitorability from 0.94 to 0.22.
- Overthinking reduces reasoning accuracy by 3.6 points.
- DiscoveryAgent identifies 14x more errors than manual review.
- SLED achieves 2,300x latency improvement over slicing.
- Readie reduces sandbox startup latency by 3–22×.
- RegimeCouncil improves accuracy by 5.2–7.3 points.
- MedGuardBench proves personalized guardrails outperform standard ones.
- PeriaTrace diagnoses teeth with 94.37% F1 accuracy.
- Agent Trust Infrastructure increases containment from 43% to 84%.
Sources
- Auditors Struggle to Catch Sabotage In End-to-End Transcripts of AI Safety Research
- Chain-of-Thought Controllability Can Reduce Chain-of-Thought Monitorability
- NEBench R1: a Reasoning Benchmark for Evaluating Nuclear Domain Competence in Large Language Models
- DiscoveryAgent: A Framework for Scientist-centric Agentic AI
- Conserved Verdicts, Drifting Commitments: Identifying Harmful Overthinking
- SLED: Sensor SLicing from Edge to Datacenter
- SkillBuffBench: Measuring What Agent Skills Contribute across Harnesses, Models, Skill Access Modes and Memory
- On Formal Counterfactual Explanations
- KV-LTM: Persistent Per-Session KV Caches as Long-Term Memory for Personal Assistants
- WebAgents: Unbundling Agent Creation, Hosting, and Execution for an Agent-Native Web
- MedGuardBench: A Patient-Oriented Benchmark for Personalized Guardrail Development
- A Hybrid Exact-Heuristic Solver for the Integrated Passenger and Parcel Transportation Problem
- Long-Context Generation Is a Sampling Problem (When You Turn Up the Heat)
- Not All Evidence Is Admissible: Evidence Governance for Tool-Augmented Medical Agents
- TeleChat to Xing4: Advancing Frontier Agentic Models on Domestic Chips and Frameworks
- RegimeCouncil: Paired-Evidence Routing For Memory Conflict Governance In LLM Agents
- From Sorry to QED: Refining LLM Math Proofs via Angelic Execution
- EvoDisco: Strategy Discovery for Evolving Agent Memory
- ATLAS-Finance: Benchmarking AI Agents in Realistic Banking Environments
- The Only Edge Was Memorized in Training: LLM Trading Agents Around Earnings and Bad News
- Readie: Low-Latency Code Sandboxes for AI Agents
- RehearsalBench: Can AI Predict What Happens When You Change a Company?
- Context Has a Price: Adding and Removing Context for LLM Coding Agents under Prefix Caching
- The Fragility of Reasoning-Importance Attribution for Pruning Language Models
- SciWorkBench: A Benchmark for Real Scientific Workflows
- FrameRouter: Frame Budget Routing for Long-Video Understanding on Video-MME-v2 and Beyond
- Towards Physics-Constrained Rectified Flow for Climate Downscaling: What Projections Can Certify?
- DACE: Diversity-Driven Adversarial Co-Evolution for Robust LLM Safety Alignment
- Fidelity Regret: Measuring Simulator Misspecification in Verifiable Design
- Streaming On-Policy Distillation: Decoupling Three Clocks for Stage-Wise Supervision Consumption
- SpecSearch: Towards Hardware-Aware Efficient Speculative Decoding
- ConfGuard: a physics-grounded reference monitor for predictive-maintenance large-language-model agents
- EviDARL: Evidence-Centric Dual-Agent Reasoning via Alternating Reinforcement Learning for Multi-Table Question Answering
- Efficient Wide-Beam Generative Retrieval with Model-Native Trie Decoding and Shared-Context Attention
- Learning to Edit IFC Building Models: Verifiable Checkers as the Training Signal for Dependency-Aware Editing Agents
- Symbolic Abstraction Refinement
- AutoE2E: Don't speak - Language free end-to-end autonomy A Real-Time Driving Model for L2++ ADAS and L4 Robotaxis
- Seeing What Still Holds: How Waiting Control Changes Cooperative Recovery
- Reliable Evaluation of AI Agents for Investment Research: A Framework for Performance and Validity Assessment
- From Ideal to Real: Integrating Language Models into Smart Home for Human-Centric IoT Safety and Security Fuzzing
- PACD: Provenance-Aware Consensus Decoding for Mitigating Cross-Modal Hallucinations in Omnimodal Large Language Models
- Proceedings of the 2026 XCSP3 Competition
- When Agent Skills Go Stale: Runtime Behavior and Selective Maintenance in Evolving Environments
- LMLineage: Content Lineage and Context Governance for LLM Agents
- DARWIN: a Geospatial Foundation Model for Marine Intelligence
- DAQ: Decision-Aligned Post-Training Quantization for Spiking Transformers
- EvoSkill-S1: Evolving Reusable Skills with Genetic-Inspired Operators
- Simplicity Goes Far: Auditing Prompt Optimizers with Demonstration-Conditioned Prompt Search
- Deep Reinforcement Learning for Heat Pump and Underfloor Heating Control: An End-to-End Case Study
- Proximal Targets: Training Looped Refiners on Problems with Many Correct Answers
- Remembered as Brief: Stored Preferences SilentlyRemove Decision-Critical Information
- Worldviews in Large Language Models
- Pluggable Subband-Decoupled Fusion and Unlabeled Capture Prototype Calibration: Bridging High-SNR Null-Subcarrier Regrowth and Low-SNR Coherent Preambles for Cross-Domain IoT RF Fingerprinting
- CrossEnv: Synthesizing Verifiable Cross-Environment Tasks for Long-Horizon Agent Training
- Can Activation Steering Substitute for Prompt-Based Affective Priming in LLM Digital Twins?
- Synergizing Evidence Retrieval and Utilization for Self-Improving Multimodal Scientific Document Reasoning
- Born for the Task: Rethinking Quantum Chemistry with Foresighted Agentic Computing
- Against Coercion: AI Alignment as a Synthetic Social Contract
- Structured AI ideation yields physical strategies for epithelial barrier repair
- VITAL: A Framework for Validating Agentic AI in Pharmaceutical Drug Development and Clinical Trial Operations
- Done, But Still Working: Anytime Evaluation of When Coding Agents Should Stop
- Residual-Guided Lightweight Reinforcement Learning for Risk-Aware Trajectory Selection with a Frozen World Model
- PeriaTrace: Tooth-Level Grounding Diagnosis and Patient-Facing Explanations from Periapical Radiographs
- Hidden-State Similarity Predicts Re-Elicitation After Inoculation Prompting
- Colombo: A Frontier-to-Frontier Autoresearch System for Oracle-Based Optimization
- The Resolution Hypothesis: Agent Evolution's Key Is Identifying Improvements
- Values Shape What Self-Improving Coding Agents Build
- Agent Trust Infrastructure (ATI): A Federated Identity, Trust, Attestation, and Accountability Architecture for Cross-Organization Autonomous AI Agents
- WoRD: Diffusion-Based Posterior Inference for Online Goal Recognition
- EDA: Observe, Verify, and Assess for Evidence-Grounded Multimodal Depression Analysis
- Right Diagnosis, Unsafe Route: Auditing Clinical AI Decision Trajectories and Corrective Oversight for Validity and Potential Harm
- MillBill: A Machining Agent with Geometric Feedback
- CADMAGEDDON: High-Fidelity CAD Reconstruction with Local Inference
- The Between-Round Decision in Autonomous Research: An LLM Controller That Reads the Whole Trajectory
- ROSMonitoring in Action: Runtime Monitoring and Enforcement for ROS 2 Actions
- HealthCLBench:Evaluating Context Learning in Health Conversations through Controlled Case Clusters
- Multistage Contextual Optimization for Vehicle Repositioning
- A Visual Information Extraction System Using Multimodal Large Language Models for Scanned Cognitive and Behavioral Assessments
- Multi-Agent Memory Inconsistency: Where Factual Conflicts Disappear in LLM Agents
- Fine-Grained Japanese Clinical Named Entity Recognition: A Controlled Comparison of BERT and Open-Weight Large Language Models
- RAG in a Trenchcoat: When Minimal Memory Is Enough for Agentic Systems, and When It Isn't
- CorrectCast: Selective Multimodal Context for Scale-Aware Residual Correction in Precipitation Nowcasting
- Reconfiguring Procedural Knowledge for Compositional Robot Skill Adaptation
- Can You Undo It? Reversibility Classes for AI Agent Actions
- Thinking Harder Doesn't Lead to Better Choices: Reasoning Effort Increases the Confidence but Not the Correctness of LLM Shopping Agents
- What Does an Agent Need? Bundle-Aware Agent Resource Composition
- RSI-Prime: Training Models to Train Models
- ForecastGym: A Live, Controlled, and Replayable Benchmark for Forecasting Agents
- Verifying Agents in Rubric-Graded Environments
- OICycle: Observe, Infer, and Intervene Cycle for Belief-State Learning in Longitudinal Clinical Decision Making
- A Full-Ledger Counterfactual Audit of Adaptive Reasoning
- MCTS Is Counting the Wrong Thing: Learning Exploration Counts for LLM Reasoning
- Harnesses That Improve Themselves: A Unifying Framework for Self-Improving Agent Harnesses
- Learning Regions of Attraction for Transverse Feedback Linearization under Controller Singularities
- SeasonRainBench: Global Seasonal Precipitation
- Honest Brokers in AI Regulation: How Public Comments Introduce Evidence to Regulators
- DNeo: Your Parallel Draft Model Needs Better Target Feature Encoding
- Language Models Can Rewrite Their Own Weights
- How Much of SAR Vehicle Recognition Is the Vehicle? Background Shortcuts, Angle Gaps and Partial Libraries on MSTAR and ATRNet-STAR
- PhysScope: Agentic Multimodal Physics Reasoning with Grounded Visual Memory
- RagHarnessDebugger: Jev-Guided Agentic RAG in Contract Inference Repair Auditing
- Outcome-Balanced PPO and Policy-Induced Flow Rectification for Flow-Based VLA Policies
- Big Brains, Tiny Board: Pupavision, an Agentic Physical AI Instrument for Monitoring Drosophila Pupariation and Lifecycle Transitions
- LoCaR: Localized Canvas Retrieval With Frozen Vision-Language Model
- When Lower Harm Misleads: Auditing Concept Corrections
- Towards Data-Efficient, Real-Time World Models
- Think Once, Act Fast: Cost-Effective Recursive Self-Improvement for Decision Model Harness
- COTTI: Contrastive Training and Time-Aware Inference for Robust Industrial Recommendation under Promotional Drift
- PACT: Plan-Action Consistency Training by Preserving Correct Plans and Repairing Actions in GUI Agents
Comments
Please log in to post a comment.