| 2026-07-23 |
ICAE-Bench: Evaluating Coding Agents as Interactive Project Builders |
— |
— |
| 2026-07-23 |
AREX: Towards a Recursively Self-Improving Agent for Deep Research |
— |
— |
| 2026-07-23 |
Show, Don't Tell: Evaluating Spatial Cognition in Generative Pixels Rather Than LLM Text |
— |
— |
| 2026-07-23 |
Streaming Multi-Agent Autoregressive Diffusion Model with World State Registers |
— |
— |
| 2026-07-23 |
OpenForgeRL: Train Harness-native Agents in Any Environment |
— |
MSR team; trains agents inside real harnesses (Claude Code, Codex); directly actionable for harness builders |
| 2026-07-23 |
PATS: Policy-Aware Training Scaffolding for Agentic Reinforcement Learning |
— |
Policy-aware scaffolding for long-horizon RL agents; addresses exploration collapse in sparse-reward settings |
| 2026-07-23 |
GuardianAgentBench: Where Agents Fail and How to Guard Them |
— |
580-scenario safety benchmark for autonomous agents; maps failure modes and guards—directly useful for harness evaluation |
| 2026-07-23 |
The Dark Room in the Reward Channel: Dense Prediction Rewards Collapse GRPO-Trained LLM Agents -- and What Actually Works |
— |
Dense reward collapse under GRPO for LLM agents; identifies what actually works—critical RL training insight |
| 2026-07-23 |
AttriMem: Attribution-Guided Process Feedback for Agent Memory Learning |
— |
Attribution-guided memory learning for agents; addresses what to store/compress/discard—core harness problem |
| 2026-07-23 |
MemTools: A Unified Research Framework for Interoperable Agent Memory |
— |
Unified interoperable agent memory framework; targets architectural fragmentation across memory lifecycle stages |
| 2026-07-23 |
Is Deep Research Reliable? Misleading Knowledge Induces False Conclusions |
— |
Reliability of deep research agents under misleading knowledge; evaluates long-horizon retrieval+synthesis pipelines |
| 2026-07-22 |
DocOps: A Verifiable Benchmark for Autonomous Agents in Complex Document Operations |
— |
— |
| 2026-07-22 |
NVIDIA-labs OO Agents: Native Python Object-Oriented Agents |
— |
— |
| 2026-07-22 |
ArbiGraph: Arbitrarily Scalable Verifiable Task Graphs for Evaluating Context Management |
— |
Scalable benchmark for context management in tool-assisted agents; evaluates retain/update/discard across workflows |
| 2026-07-22 |
From Agent Failures to Text Policies: What Works and What Breaks |
— |
Empirical analysis of TextGrad applied to agents; identifies what breaks in text-gradient optimization for agent components |
| 2026-07-21 |
Cross-Agent Campaign Attribution: Linking Asynchronous Attacks Across LLM Agents |
ICML 2026 |
ICML 2026; first formalization of distributed cross-agent cross-session attack attribution |
| 2026-07-21 |
ResearchArena: Evaluating Sabotage and Monitoring in Automated AI R&D |
— |
AI-control approach to untrusted R&D agents; sabotage detection + monitoring benchmark |
| 2026-07-21 |
PhoenixRepair: Rethinking Repair Strategy Exploration in Software Agents |
— |
Daya Guo (DeepSeek); novel repair-strategy exploration for software-agent issue resolution |
| 2026-07-21 |
They'll Verify. They Just Won't Act. How Authority Framing and Laundered Code Turn a Trusted Agentic CI/CD Pipeline Into an Attack Surface |
— |
Production 5-agent CI/CD pipeline; authority-laundering turns guardrails into attack surface |
| 2026-07-21 |
AgentDebugX: An Open-Source Toolkit for Failure Observability, Attribution, and Recovery in LLM Agents |
— |
Open-source toolkit for LLM agent failure attribution and root-cause recovery |
| 2026-07-21 |
MedDDC-Eval: Diagnosis-Decoupled Evaluation of Multi-Turn Medical Consultation Agents |
— |
Decouples inquiry policy from diagnosis quality in multi-turn medical consultation agents |
| 2026-07-21 |
AgentTrails: Towards Trust and Reuse for Agentic Tasks |
— |
Freire group (NYU); provenance and trust framework for reusable agent task trajectories |
| 2026-07-21 |
Broken Gates: Re-evaluating Web Bot Defenses in the Age of LLM Agents |
— |
LLM agents circumvent CAPTCHAs and bot defenses; forces web-security threat-model update |
| 2026-07-21 |
Supra Cognitive Modes: A Routed Architecture for Agent Memory |
— |
Routed memory architecture separating factual lookup, reasoning, synthesis — core harness primitive |
| 2026-07-21 |
CodeRescue: Budget-Calibrated Recovery Routing for Coding Agents |
— |
Budget-calibrated failure-recovery routing for coding agents; practical cost-aware scaffolding mechanism |
| 2026-07-21 |
Data Leakage Prevention in Agentic Applications via Preemptive Hardening |
— |
Preemptive hardening against prompt-injection data leakage; actionable security for agentic deployment |
| 2026-07-21 |
Agents in the Wild: Where Research Meets Deployment |
— |
Research-to-production survey from credible IR/agent group; maps real deployment failure modes |
| 2026-07-21 |
Verifiable Self-Evolution for Open-Ended Dialogue Skills via Future-Feedback Prediction |
— |
Self-evolution for frozen agents via future-feedback; extends verifiable RL to open dialogue skills |
| 2026-07-21 |
SkillSight: Seeing Through Shared Descriptions for Accurate Skill Retrieval |
— |
Skill retrieval for large libraries; solves accuracy bottleneck in skill-augmented agent harnesses |
| 2026-07-21 |
Skillware: A Software Ontology and Engineering Lifecycle for Persistent Behavioral Artifacts |
— |
First software ontology treating agent skills as persistent artifacts with lifecycle and metadata |
| 2026-07-21 |
BioSecBench-Surveillance: A Verifiable Benchmark for AI Agents in Pathogen Genomic Surveillance |
— |
Verifiable benchmark for AI agents in pathogen surveillance; medical AI + agent eval combined |
| 2026-07-21 |
FinanceComplexQA: Benchmarking Agentic Reasoning on Industrial-grade Financial Documents |
— |
— |
| 2026-07-20 |
ETAS: An Effect-Typed Language for Agent Systems |
— |
First-class language semantics for agents, tools, memory, policies — foundational harness design |
| 2026-07-20 |
Harness Engineering for LLM-Driven GPU Kernel Generation |
— |
Harness-centered LLM agent system for GPU kernels; constraint, validate, profile, select loop |
| 2026-07-20 |
TRIM: Reducing AI-Generated CodeSlop via Agent Trajectory Minimization |
— |
Google Research; trajectory minimization cuts agent-generated code bloat — practical coding agent fix |
| 2026-07-20 |
Verify, Repair, Repeat, or Stop? Robust Stopping for Noisy Verify-Repair Loops in LLM Agents |
— |
Formal stopping criteria for noisy verify-repair loops; directly governs agent loop design |
| 2026-07-20 |
FlashRT: Agent Harness for Guiding Agents to Deploy Real-Time Multimodal Applications |
— |
Agent harness for real-time multimodal pipelines; placement, streaming, intra-pipeline decisions |
| 2026-07-20 |
SEE: Structure-aware Exploring \& Exploiting for Long-horizon GUI Agent Trajectory Synthesis |
— |
Long-horizon GUI agent trajectory synthesis; structure-aware explore/exploit for scarce data |
| 2026-07-20 |
MADA-RL: Multi-Agent Debate-Aware Reinforcement Learning for Parameter-Efficient Reasoning in Compact Models |
TMLR |
TMLR; Tresp group; multi-agent debate + RL trains compact models to reason efficiently |
| 2026-07-20 |
Adaptive Adversaries: A Multi-Turn, Multi-LLM Benchmark for LLM Agent Security |
— |
Adaptive multi-turn, multi-LLM security benchmark; dynamic attacker vs. defender evaluation |
| 2026-07-20 |
SWE-Pruner Pro: The Coder LLM Already Knows What to Prune |
— |
Microsoft Research; coding agent self-prunes context without separate classifier — deploy-ready |
| 2026-07-20 |
Mechanistic Attention Guidance for Agent Memory Refinement |
— |
Mechanistic attention signals improve agent memory refinement beyond text-only trajectories |
| 2026-07-20 |
Why Does Feedback-Augmented Self-Distillation Fail to Improve Retrieval-Interleaved Search Agents? |
— |
Negative result on OPSD for retrieval-interleaved agents; tells builders what training strategies fail |
| 2026-07-20 |
Self-State Attacks on Self-Hosted AI Agents: How Far Can OS Defenses Go? |
— |
Schmidhuber co-author; self-state OS-level attack vector on self-hosted agents is novel and underexplored |
| 2026-07-20 |
Is Progressive Disclosure All You Need for Long-Context Agents? |
— |
Benchmarks progressive-disclosure vs retrieval for long-context agents; direct harness architecture guidance |
| 2026-07-20 |
Oracle Gap and Signal Fidelity: A Fixed-Pool Diagnostic for Test-Time Collaboration |
— |
Fixed-pool diagnostic separates oracle gap from signal fidelity; clarifies when verifier pipelines help |
| 2026-07-20 |
Retain or Consolidate? Budget-Dependent Operator Selection for Language Agent Memory |
— |
Budget-aware retain-vs-consolidate memory operator; directly addresses context-cost tradeoffs in agents |
| 2026-07-20 |
Exploratory and Assimilating Reflection: Reflective Recall Cycle for Long-term Memory |
— |
Reflective recall cycle for long-term agent memory; novel two-phase (explore + assimilate) design |
| 2026-07-20 |
The Shared Discovery Paradox: How a One-Answer Rule Turns Better Information into Worse Search |
— |
Nakajima (BabyAGI); exactly solvable multi-agent shared-discovery model exposes coverage vs belief tradeoff |
| 2026-07-20 |
ProEvent: An Event-centric Benchmark for Proactive Agents |
— |
Event-centric proactive agent benchmark; evaluates anticipatory assistance without explicit user instructions |
| 2026-07-20 |
Structured Output Collapses Answer Diversity Across 44 Language Models |
— |
Structured-output format shifts answers across 44 LLMs; critical for every agent builder |
| 2026-07-20 |
The Story Shapes the Agent: Narrative Priors in LLM Behavior |
COLM 2026 |
COLM 2026; narrative framing empirically dominates persona in steering agent behavior |
| 2026-07-20 |
Engineering Trustworthy Agentic AI for Critical Systems |
— |
Engineering framework for trustworthy agentic AI in critical systems; evaluation and safety criteria |
| 2026-07-18 |
DataFlow-Harness: A Grounded Code-Agent Platform for Constructing Editable LLM Data Pipelines |
— |
Novel platform closing script-vs-artifact gap; directly models editable agent harness pipelines |
| 2026-07-17 |
Recursive Harness Self-Improvement |
— |
Zaharia (Berkeley/Databricks); harnesses as data-generators that self-improve — fundamental for meta-harness builders |
| 2026-07-17 |
When Do Multi-Agent Systems Help? An Information Bottleneck Perspective |
— |
Information-bottleneck theory for when MAS beats SAS; directly governs single- vs multi-agent architecture decisions |
| 2026-07-17 |
Process Reward Informed Tree Rollout for Effective Multi-Turn RL |
— |
Process-reward tree rollout for multi-turn RL; solves GRPO/RLOO credit-assignment failure on long-horizon agent tasks |
| 2026-07-17 |
ToolVerse: Unlocking Massive Environments and Long-Horizon Tasks for Agentic Reinforcement Learning |
— |
ToolVerse: massive diverse environments + long-horizon tasks close the evaluation gap for real-world agentic RL |
| 2026-07-17 |
SeerGuard: A Safety Framework for Mobile GUI Agents via World Model Prediction |
— |
World-model safety predictor for GUI agents before irreversible actions; directly applicable architecture for any action-space agent |
| 2026-07-17 |
DSWorld: A Data Science World Model for Efficient Autonomous Agents |
— |
World model cuts expensive trial-and-error in data-science agents; pattern generalizes to any code-execution agent loop |
| 2026-07-17 |
SkillCorpus: Consolidating and Evaluating the Open Skill Ecosystem for Real-World LLM Agents |
— |
First corpus + eval of SKILL.md ecosystem; ground truth on skill quality and reuse for real-world agent harnesses |
| 2026-07-17 |
Scalable LLM Agent Tool Access in the Cloud |
— |
MCP at cloud scale; concrete infra for tool-calling agents — immediately buildable |
| 2026-07-17 |
Behavioral Controllability of Agentic Models for Information Extraction: From Fixed Workflows to Reflective Agents |
— |
Empirically tests whether reflection/memory in LLM agents yield controllable, observable gains |
| 2026-07-17 |
Do Agents Dream of False Memories? Black-box Visual Attacks on Long-term Memory in Multimodal AI Agents |
— |
Black-box memory poisoning attacks on multimodal agents; critical security surface for long-term memory systems |
| 2026-07-17 |
Agentic Synthesis against Counterexample-Supplemented Sketches |
— |
Counterexample-supplemented synthesis prevents coding agents from repeating domain-rule violations |
| 2026-07-17 |
The Honest Quorum Problem: Epistemic Byzantine Fault Tolerance for Agentic Infrastructure |
— |
Byzantine fault tolerance framing for agentic infrastructure; novel epistemic trust model |
| 2026-07-17 |
When Does Muon Help Agentic Reinforcement Learning? |
— |
Muon vs AdamW in sparse-reward agentic RL; optimizer choice has direct post-training implications |
| 2026-07-17 |
Knowledge-Centric Agents for Workflow Generation |
ECCV 2026 |
ECCV 2026; knowledge-centric agents for workflow generation in visual creation systems (ComfyUI) |
| 2026-07-17 |
RecGPT-V3 Technical Report |
— |
— |
| 2026-07-16 |
Democratizing Agent Deployment Safety: A Structural Monitoring Approach |
ICML 2026 |
ICML 2026; structural runtime monitoring for deployed agents; directly buildable harness component |
| 2026-07-16 |
Bad Memory: Evaluating Prompt Injection Risks from Memory in Agentic Systems |
— |
Franziska Roesner (UW security); memory-as-attack-surface; critical threat model for persistent agents |
| 2026-07-16 |
SEED: Self-Evolving On-Policy Distillation for Agentic Reinforcement Learning |
— |
Self-evolving on-policy distillation closes RL sample-efficiency gap; directly improves agentic training recipes |
| 2026-07-16 |
Proof-or-Stop: Don't Trust the Agent, Trust the Evidence -- Loop Engineering for Verifiable Evidence-Gated Lifecycle Control |
— |
Evidence-gated lifecycle control; novel harness primitive for verifiable agent state transitions |
| 2026-07-16 |
Bridge Evidence: Static Retrieval Utility Does Not Predict Causal Utility in Multi-Step Agentic Search |
— |
Static retrieval utility ≠ causal utility in multi-step search; redefines how to build agentic RAG |
| 2026-07-16 |
MCPEvol-Bench: Benchmarking LLM Agent Performance Across Dynamic Evolutions of MCP Servers |
— |
MCP is now core agent infrastructure; first benchmark tracking agent perf across server version drift |
| 2026-07-16 |
MemPoison: Uncovering Persistent Memory Threats and Structural Blind Spots in LLM Agents |
— |
Memory poisoning via normal interaction channels; persistent cross-session attack surface with concrete exploits |
| 2026-07-16 |
Tactile: Giving Computer-Using Agents Hands and Feet |
— |
Replaces brittle coordinate-click motor layer for computer-use agents; practical foundation for GUI automation |
| 2026-07-16 |
LongStraw: Long-Context RL Beyond 2M Tokens under a Fixed GPU Budget |
— |
Closes RL/inference context gap (2M tokens); novel approach for long-horizon agent training |
| 2026-07-16 |
Plover: Steering GUI Agents through Plan-Centric Interaction |
— |
Plan-centric GUI agent handles dynamic layouts; core capability for building robust GUI agents |
| 2026-07-16 |
Does Multi-Agent Debate Improve AI Feedback on Research Papers? |
— |
Pre-registered negative result: multi-agent debate doesn't improve AI feedback; critical harness design signal |
| 2026-07-16 |
RetroAgent: Harnessing LLMs to Search Over Structured Memory for Agentic Retrosynthesis Planning |
COLM 2026 |
COLM 2026; structured memory for LLM agentic planning in complex combinatorial search spaces |
| 2026-07-16 |
AutoSynthesis: An agentic system for automated meta-analysis |
— |
Feuerriegel (TU Munich); end-to-end agentic scientific synthesis; reproducible for medical evidence automation |
| 2026-07-16 |
Beyond Generalist LLMs: Specialist Agentic Systems for Structured Code Workflow Execution |
— |
Specialist vs generalist coding agents; structured workflow execution beats general LLMs on complex tasks |
| 2026-07-16 |
When Words Are Safe But Actions Kill: Probing Physical Danger Beyond Text Safety in Hidden-State Risk Space |
— |
LLM-as-planner safety gap: linguistically safe → physically lethal; critical for embodied agent harnesses |
| 2026-07-16 |
OmniaBench: Benchmarking General AI Agents Across Diverse Scenarios |
— |
Broad general-agent benchmark spanning tool use + multi-step tasks; key eval reference for harness builders |
| 2026-07-16 |
SearchOS-V1: Towards Robust Open-Domain Information-Seeking Agent Collaboration |
— |
Multi-agent web-search collaboration addressing long-context task tracking; directly buildable system |
| 2026-07-16 |
Step-Level Preference Learning for Generative Agents in Social Simulations |
— |
Step-level preference learning for long-horizon agent decisions; advances beyond episode-level RLHF for agents |
| 2026-07-16 |
SafeRelBench: A Spatial-Relation-Aware Benchmark for Process-Level Safety in VLM-Driven Embodied Agents |
— |
Process-level spatial-relation safety benchmark for VLM embodied agents; fills eval gap for robot planners |
| 2026-07-16 |
TopoAgent: A Self-Evolving Topological Agent for Multimodal Scientific Reasoning |
— |
Self-evolving topological agent for multimodal scientific reasoning; novel non-linear planning architecture |
| 2026-07-16 |
MM-IssueLoc: A Controlled Benchmark for Evaluating Visual Evidence in Multimodal Repository-Level Issue Localization |
— |
Multimodal repo-level issue localization benchmark; exposes text-only gap in coding agent eval |
| 2026-07-16 |
Are LLM-Generated GPU Kernels Production-Ready? A Trace-Driven Benchmark and Optimization Agent |
— |
Trace-driven production benchmark + optimization agent for LLM-generated GPU kernels; practical infra-level eval |
| 2026-07-16 |
BrainPilot: Automating Brain Discovery with Agentic Research |
— |
End-to-end agentic pipeline for neuroscience discovery; strong template for scientific agent harness design |
| 2026-07-16 |
ANet Patu-1: The Value of Connection in the Agent Network |
— |
Novel network-value theory (Reed/Metcalfe laws) applied to multi-agent topology design |
| 2026-07-16 |
Scaling Behavior Foundation Model for Humanoid Robots |
— |
Scaling behavior foundation model for humanoid embodied agents; foundational for generalist robots |
| 2026-07-16 |
VTM-Nav: Hierarchical Visual-Topological Memory for Cross-Episode Object-Goal Navigation |
— |
Training-free VLM-based embodied navigation with hierarchical visual-topological memory |
| 2026-07-16 |
Digital Pantheon: Simulating and Auditing Coalition Formation with LLM Agents |
— |
Multi-LLM agent coalition simulation + auditing framework; practical multi-agent behavior evaluation |
| 2026-07-16 |
MathCoPilot: An Interactive System for Human-AI Symbiotic Paradigm of Mathematical Research |
— |
Human-AI symbiotic paradigm for math; novel human-in-loop agent scaffold for scientific discovery |
| 2026-07-16 |
Alipay-PIBench: A Realistic Payment Integration Benchmark for Coding Agents |
— |
Industry coding-agent benchmark (Alipay) covering realistic repo-level payment integration tasks |
| 2026-07-16 |
The Energy Society: A Simulation Environment for Studying Agent Cooperation under Survival Pressure |
— |
Controlled multi-agent survival environment to study emergent cooperation/defection in LLM agents |
| 2026-07-16 |
Towards an Intention Abstraction Layer for Autonomous Industrial Systems |
— |
Intention abstraction layer for multi-subsystem coordination; directly relevant to agent harness design |
| 2026-07-16 |
StructureClaw: Traceable LLM Agents and an Executable Benchmark for Structural Engineering Workflows |
— |
Traceable LLM agents + executable benchmark for multi-step engineering workflows |
| 2026-07-16 |
LQCDMaster: Agentic Scientific Computing for Lattice Quantum Chromodynamics Research |
— |
End-to-end agentic scaffold for scientific computing; tests multi-step tool orchestration |
| 2026-07-16 |
Collaborative Spatial Learning with Multi-LLM Agents in Networked Social Experiments |
— |
Multi-LLM agents in networked experiments; exploit/explore tradeoff across agent collectives |
| 2026-07-16 |
Memory-Driven Self-Disclosure and Relational Turning Points: A Longitudinal Multimodal Study of Human-AI Interaction |
— |
Longitudinal memory-augmented conversational agent; empirical relational dynamics data |
| 2026-07-16 |
Can We Trust Item Response Theory for AI Evaluation? |
— |
Questions IRT foundations of AI benchmarking; directly impacts agent evaluation validity |
| 2026-07-16 |
SAGA: Schema-Aware Grounding for Agentic Text-to-SPARQL Generation |
— |
Schema-aware grounding for agentic SPARQL; structured KG query generation via LLMs |
| 2026-07-16 |
MARS: Multi-hop Adaptive Retrieval and SPARQL Generation for KGQA |
— |
Multi-hop adaptive retrieval + SPARQL; grounded LLM reasoning over knowledge graphs |
| 2026-07-16 |
Multimodality as Supervision: Self-Supervised Specialization to the Test Environment via Multimodality |
— |
Self-supervised cross-modal specialization at test time; practical for embodied VLM agents |
| 2026-07-16 |
SciDiagramEdit: Learning to Edit Scientific Diagrams from Paper Revisions |
— |
Figure editing via paper revision diffs; novel multimodal grounding for document agents |
| 2026-07-16 |
Concept-Guided Spatial Regularization for World Models in Atari Pong |
— |
— |
| 2026-07-16 |
Reachability-Aware Pretraining for Efficient Target-Oriented Path Exploration in Temporal Knowledge Graph Reasoning |
— |
— |
| 2026-07-16 |
RW-Voice-EQ Bench: A Real World Benchmark for Evaluating Voice AI Systems |
— |
— |
| 2026-07-16 |
Dialogue Summarization with Emotion Dynamics Using Topic- and Participant-Centric Decomposition |
— |
— |
| 2026-07-16 |
An Intelligent-Cloud Edge Multimodal Interaction System for Robots |
— |
— |
| 2026-07-16 |
Knowing You at First Glance: Inferring Apparent Personality from Faces |
— |
— |
| 2026-07-16 |
Precise but Uncoupled: Reviewer Precision Does Not Guarantee Critique Uptake in Multi-Agent Math Reasoning |
— |
4k+ example study proving reviewer precision decoupled from critique uptake — breaks a core MAS design assumption |
| 2026-07-16 |
Coercion and Deception in AI-to-AI Management: An Agentic Benchmark of Unprompted Escalation |
— |
First benchmark of unprompted coercion/deception in AI-to-AI authority chains; critical safety signal for orchestrators |
| 2026-07-16 |
Beyond Success Rate: Cost-Aware Evaluation of Offensive and Defensive Security Agents |
— |
Singer (Harvard); reframes security-agent eval as cost-aware rather than peak-capability — changes how benchmarks are designed |
| 2026-07-16 |
Do Coding Agents Need Executable World Models, Simplification, and Verification to Solve ARC-AGI-3? |
— |
Ablates ARC-AGI-3 agent components (world model, simplification, verification); direct scaffold design signal |
| 2026-07-15 |
Early Adoption of Agentic Coding Tools by GitHub Projects |
KDD 2026 |
— |
| 2026-07-15 |
Multi-Agent Collaborative Reasoning with Tool-Augmented Evidence for Urban Region Profiling |
KDD 2026 |
— |
| 2026-07-15 |
AI-accelerated End-to-End Framework for Rapid Professional Upskilling |
— |
— |
| 2026-07-15 |
Rethinking Penetration Testing for AI-Enabled Systems: From Resource Compromise to Behavioral Objective Violation |
— |
— |
| 2026-07-15 |
Do Agent Optimizers Compound? A Continual-Learning Evaluation on Terminal-Bench 2.0 |
— |
— |
| 2026-07-15 |
Lyapunov Exponent as Physics-Informed Dense Reward: RL Discovery of Stabilization Beyond the Kapitza Pendulum |
— |
— |
| 2026-07-15 |
The Dynamic Verifiable Multi-Agent Human Agentic Loyalty Loop (DVM-HALL) Model and the Net Human-Agent Score (NHAS) in Autonomous Commerce |
— |
— |
| 2026-07-15 |
TRACE: Turn-level Reward Assignment via Credit Estimation for Long-Horizon Agents |
— |
— |
| 2026-07-15 |
A Self-Evolving Agent for Longitudinal Personal Health Management |
— |
— |
| 2026-07-15 |
DeepStress: Stress-Testing Deep Search Agents |
— |
— |
| 2026-07-15 |
Self-Evolving Agent Harnesses via Gated Semantic Quality-Diversity |
— |
Self-evolving harness via quality-diversity; directly buildable; rare harness-first framing |
| 2026-07-15 |
Memory as a Controlled Process: Learned Adaptive Memory Management for LLM Agents |
— |
Learned adaptive memory access (not fixed retrieval); novel control-process framing for agents |
| 2026-07-15 |
AgentCompass: A Unified Evaluation Infrastructure for Agent Capabilities |
— |
Unified, decoupled evaluation infra; addresses reproducibility gap across agent benchmarks |
| 2026-07-15 |
Set-shifting Behavioral Test for Harnessed Agents |
— |
Cognitive-psychology set-shifting to test harness adaptation to silent tool-reliability shifts |
| 2026-07-15 |
SPyCE: Skill-Policy Co-evolution for Multimodal Agents |
— |
RL skill-policy co-evolution for multimodal tool-using agents; addresses sparse reward problem |
| 2026-07-15 |
CAVA: Canonical Action Verification and Attestation for Runtime Governance of Agentic AI Systems |
— |
Cross-runtime action attestation; governance layer for heterogeneous agentic deployments |
| 2026-07-15 |
LAPO: Leave-One-Turn Attribution for Self-Generated Process Rewards in Multi-Turn Search Reasoning |
— |
Self-generated process rewards for multi-turn search; removes need for dense human annotation |
| 2026-07-15 |
Reward-Free Evolving Agents via Pairwise Validator |
— |
Reward-free self-evolution via pairwise validator; removes costly reward engineering from agent improvement loops |
| 2026-07-15 |
KnowAct-GUIClaw: Know Deeply, Act Perfectly, Personal GUI Assistant with Self-Evolving Memory and Skill |
— |
Self-evolving GUI agent with memory and skill accumulation; cross-platform deployment relevance |
| 2026-07-15 |
ABot-AgentOS: A General Robotic Agent OS with Lifelong Multi-modal Memory |
— |
General robotic agent OS with lifelong multimodal memory; runtime architecture for agent harnesses |
| 2026-07-15 |
CausalGraphX: A Counterfactual Graph Neural Network Framework for Explainable Systemic Risk Assessment |
AAAI 26 |
— |
| 2026-07-15 |
Decision Making Needs Uncertainty Quantification [Lecture Notes] |
— |
— |
| 2026-07-15 |
Cura 1T: Specialized Model for Agentic Healthcare |
— |
1T-param agentic healthcare model covering reasoning + workflow execution; applied medical AI |
| 2026-07-15 |
Diagnosing and Calibrating Tool-Call Boundary Drift in Multi-Teacher On-Policy Distillation |
— |
Diagnoses tool-call boundary drift in multi-teacher distillation; novel problem for agentic LM training |
| 2026-07-14 |
Evidence-Grounded Verified Agentic Reasoning: A Path Toward Eliminating LLM Hallucination in Empirical Inference via Tool-Attested Kernel Proofs |
ICML 2026 |
ICML 2026; formal evidence-attestation for agentic reasoning; directly addresses hallucination in tool-using agents |
| 2026-07-14 |
Speculate with Memory: Lossless Acceleration for LLM Agents |
— |
Chien-Sheng Wu (Salesforce AI); memory-augmented speculative execution gives lossless agent speedup; immediately buildable |
| 2026-07-14 |
Tracing Agentic Failure from the Flow of Success |
— |
Sharon Li (UW-Madison); causal failure attribution in LLM agent trajectories; critical for harness debugging |
| 2026-07-14 |
Function-Aware Fill-in-the-Middle as Mid-Training for Coding Agent Foundation Models |
— |
Fill-in-the-Middle mid-training teaches models to handle tool returns; novel foundation for coding agents |
| 2026-07-14 |
PM-Bench: Evaluating Prospective Memory in LLM Agents |
COLM 2026 |
COLM 2026; Saadia Gabriel; first rigorous benchmark for prospective memory in long-horizon agents |
| 2026-07-14 |
Rethinking the Evaluation of Harness Evolution for Agents |
— |
Exposes train/eval leakage in harness evolution benchmarks; reshapes how agent harness progress is measured |
| 2026-07-14 |
Who Grades the Grader? Co-Evolving Evaluation Metrics and Skills for Self-Improving LLM Agents |
— |
Self-improving agents that co-evolve their own evaluation metrics; solves the hidden-metric-assumption problem |
| 2026-07-14 |
Track, Rank, Crack: Epistemic Working Memory Scales Multi-Hop Reasoning in Language Agents |
— |
Epistemic working memory structure explicitly tracks confirmed/open sub-goals; scales multi-hop agent reasoning |
| 2026-07-14 |
Self-Improvements in Modern Agentic Systems: A Survey |
— |
First comprehensive survey on self-improving deployed agents; maps controllable evolution paradigms |
| 2026-07-14 |
Harness Handbook: Making Evolving Agent Harnesses Readable,Navigable, and Editable |
— |
Practical handbook for navigating/editing evolving harnesses; engineering-focused, rare artifact |
| 2026-07-14 |
Do AI Agents Know When a Task Is Simple? Toward Complexity-Aware Reasoning and Execution |
— |
Complexity-aware execution; reduces wasted tokens/steps in agent workflows, directly applicable |
| 2026-07-14 |
PalmClaw: A Native On-Device Agent Framework for Mobile Phones |
— |
Native on-device LLM agent framework for mobile; infrastructure-level, enables edge deployment of agent harnesses |
| 2026-07-14 |
Navigating the Mirage: A Dual-Path Agentic Framework for Robust Misleading Chart Question Answering |
— |
Dual-path agentic VLM framework for misleading chart QA; concrete scaffold pattern to borrow |
| 2026-07-14 |
Boogu-Image-0.1: Boosting Open-Source Unified Multimodal Understanding and Generation |
— |
Open-source unified multimodal model family; strong VLM agent backbone baseline |
| 2026-07-14 |
From Human-Centric to Agentic Code Review: The Impact of Different Generations of Generative AI Technology on Review Quality |
— |
Empirical study of AI-driven vs human code review across gen-AI generations; quality benchmarking |
| 2026-07-13 |
SCALECUA: Scaling Computer Use Agents with Verifiable Task Synthesis and Efficient Online RL |
— |
Xiao Liu (AgentBench) group; RLVR + verifiable task synthesis for computer-use agents at scale |
| 2026-07-13 |
Multi-Agent LLMs Fail to Explore Each Other |
— |
Sharon Li lab; empirical proof LLM agents fail exploration — undermines multi-agent reliability assumptions |
| 2026-07-13 |
Compile, Then Page: Executable SOP Programs and a Capability-Gated Runtime for Procedural LLM Agents |
— |
SOPs compiled to executable pseudo-code + capability-gated stack runtime; novel procedural-agent architecture |
| 2026-07-13 |
Agentic Routing: The Harness-Native Data Flywheel |
— |
Positions routing as a harness-native data flywheel; directly actionable for agent harness builders |
| 2026-07-13 |
STAMP: Provenance-Guided Credit Assignment for Deep Search Agents |
— |
Provenance-guided credit assignment for deep-search agents; novel RL signal beyond trajectory-level scoring |
| 2026-07-13 |
StructAgent: Harness Long-horizon Digital Agents with Unified Causal Structure |
— |
Causal structure for long-horizon digital agents; addresses context drift in real-world computer-use tasks |
| 2026-07-13 |
ToFu: A White-Box, Token-Efficient Agent Harness for Researchers |
— |
White-box agent harness paper; directly buildable; token-efficiency and orchestration transparency |
| 2026-07-13 |
When Local Monitors Miss Compositional Harm: Diagnosing Distributed Backdoors in Multi-Agent Systems |
— |
Exposes fundamental hole in per-step MAS monitors; critical for safe multi-agent deployment |
| 2026-07-13 |
Agent Hacks Agent: Autoresearch for Production-Agent Red-Teaming |
— |
Autoresearch red-teaming of production agents (Claude Code, Codex); automated, evolving attack pipeline |
| 2026-07-13 |
AgentCheck: A Reproduce-Intervene-Mitigate Workbench for LLM Agents over MCP |
— |
MCP-integrated reproduce-intervene-mitigate workbench; fills critical gap in agent failure tooling |
| 2026-07-13 |
A Formal Hierarchical Architecture for Agentic Orchestration with Stack-Based Execution and Lazy Discovery |
— |
Formal stack-based orchestration + lazy tool discovery; directly applicable to harness architecture |
| 2026-07-13 |
MM-ToolSandBox: A Unified Framework for Evaluating Visual Tool-Calling Agents |
— |
ECCV 2026; 500+-tool stateful benchmark for visual tool-calling agents; strong eval coverage |
| 2026-07-13 |
The Hidden Footprint: Making Storage a First-Class Metric for LLM Agent Evaluation |
— |
Introduces storage footprint as agent eval metric; novel axis overlooked by all current benchmarks |
| 2026-07-13 |
The Path to Self-Evolving Clinical Systems: Scaling Medical Agents from Assistance to Autonomy |
— |
Roadmap for medical agent autonomy scaling; bridges VLM reasoning to clinical deployment |
| 2026-07-13 |
Mako: A Self-Evolving Agentic Operating System (SE-AOS) for Autonomous Web Exploitation |
— |
Self-evolving agent OS that patches its own exploit kernel at runtime; novel self-improvement paradigm |
| 2026-07-13 |
Agentic Skill Optimization over Lie Algebroids |
— |
Mahadevan (UMass); formal Lie algebroid framework for structured skill editing in agentic systems |
| 2026-07-13 |
Token Reduction Is Not Cost Reduction |
— |
Shows token reduction doesn't equal API cost reduction; counterintuitive result every agent-builder must know |
| 2026-07-12 |
Towards Autonomous and Auditable Medical Imaging Model Development |
— |
LLM agents autonomously run medical imaging ML pipelines; directly relevant to applied medical AI + agents |
| 2026-07-11 |
GRASP: GRanularity-Aware Search Policy for Agentic RAG |
— |
Adobe Research; granularity-aware search policy directly improves agentic RAG decision-making in harnesses |
| 2026-07-10 |
Failure as a Process: An Anatomy of CLI Coding Agent Trajectories |
— |
Earl T. Barr (UCL); empirical anatomy of coding-agent failure modes — directly actionable for harness builders |
| 2026-07-10 |
KV-PRM: Efficient Process Reward Modeling via KV-Cache Transfer for Multi-Agent Test-Time Scaling |
— |
KV-cache transfer for process reward models cuts TTS cost in multi-agent systems; novel efficiency mechanism |
| 2026-07-10 |
Scoped Verification for Reliable Long-Horizon Agentic Context Evolution under Distribution Shift |
— |
Scoped verification of mutable agentic context under distribution shift — core reliability problem for deployed harnesses |
| 2026-07-10 |
ARCANA: A Reflective Multi-Agent Program Synthesis Framework for ARC-AGI-2 Reasoning |
— |
Reflective multi-agent program synthesis on ARC-AGI-2; hard public benchmark with constrained test-time compute |
| 2026-07-10 |
Shared Selective Persistent Memory for Agentic LLM Systems |
— |
Shared selective persistent memory solves session-zero cold-start in multi-turn tool-using agents; immediately buildable |
| 2026-07-10 |
Inside the Skill Market: From Software Engineering Activities to Reusable Agent Skills |
— |
Jialun Cao (HKUST); formalizes reusable agent skills from SE activities — modular agent composition architecture |
| 2026-07-10 |
ProofCouncil: An LLM Agent for Solving Open Mathematical Problems |
— |
ETH Zurich; Planner-Worker-Verifier agent for open math problems; strong group on reasoning + formal methods |
| 2026-07-10 |
LongMedBench: Benchmarking Medical Agents for Long-Horizon Clinical Decision-Making |
MICCAI 2026 |
MICCAI 2026; EHR-based long-horizon clinical agent benchmark — fills gap in medical agent evaluation beyond short-context QA |
| 2026-07-09 |
Provably Optimal Learning Algorithms for Assistance Games |
— |
Jordan + Stuart Russell (CHAI); provably optimal algorithms for assistance/alignment games |
| 2026-07-09 |
Persuasion Attacks Can Decrease Effectiveness of CoT Monitoring |
— |
DeepMind safety (Krakovna); persuasion bypasses CoT monitoring — breaks a key safety assumption |
| 2026-07-09 |
TTHE: Test-Time Harness Evolution |
— |
Test-time harness evolution; agent behavior shaped by harness, not just model weights — novel framing |
| 2026-07-09 |
Prismata: Confining Cross-Site Prompt Injection in Web Agents |
— |
Berkeley (Raluca Popa); principled confinement of cross-site prompt injection in web agents |
| 2026-07-09 |
Remember When It Matters: Proactive Memory Agent for Long-Horizon Agents |
— |
Microsoft Research (Bo Peng); proactive memory surfacing for long-horizon agent trajectories |
| 2026-07-09 |
Tool-Making and Self-Evolving LLM Agents in Low-Latency Systems |
— |
Self-evolving agents compile SOP steps into reusable tools, cutting per-request latency |
| 2026-07-09 |
SMetric: Rethink LLM Scheduling for Serving Agents with Balanced Session-centric Scheduling |
— |
Session-centric LLM scheduling rethinks serving for agents vs. human request patterns |
| 2026-07-09 |
CausalDS: Benchmarking Causal Reasoning in Data-Science Agents |
— |
— |
| 2026-07-09 |
WebSwarm: Recursive Multi-Agent Orchestration for Deep-and-Wide Web Search |
— |
WebSwarm recursive multi-agent orchestration for deep web search; directly buildable |
| 2026-07-09 |
What to Keep, What to Forget: A Rate--Distortion View of Memory Compaction in LLMs and Agents |
— |
Rate-distortion theory for agent memory compaction; principled framework for KV/context management |
| 2026-07-09 |
Cognitive-structured Multimodal Agent for Multimodal Understanding, Generation, and Editing |
— |
Cognitive-structured multimodal agent with selective context; addresses VLM context scaling |
| 2026-07-09 |
From Prompts to Contracts: Harness Engineering for Auditable Enterprise LLM Agents |
— |
Harness engineering for auditable enterprise LLM agents; contracts+source boundaries for production |
| 2026-07-09 |
ProjAgent: Procedural Similarity Retrieval for Repository-Level Code Generation |
— |
ProjAgent procedural similarity retrieval for repo-level codegen; practical agent tool-use |
| 2026-07-09 |
GitLake: Git-for-data for the agentic lakehouse |
— |
GitLake: Git-for-data agentic lakehouse; branching/versioning primitives for agent workflows |
| 2026-07-09 |
TRACE: A Two-Channel Robust Attribution Watermark via Complementary Embeddings for LLM-Agent Trajectories |
— |
Watermarking LLM-agent trajectories for provenance; security primitive for agent systems |
| 2026-07-09 |
Token-Flow Firewall: Semantic Runtime Auditing for Persistent AI Agents |
— |
Token-Flow Firewall: semantic runtime auditing for persistent agents — directly addresses agent safety/security in long-lived systems |
| 2026-07-09 |
UniClawBench: A Universal Benchmark for Proactive Agents on Real-World Tasks |
— |
UniClawBench: universal benchmark for proactive agents on real-world tasks — fills evaluation gap for agentic systems |
| 2026-07-09 |
Playing ZendoWorld: Challenging AI Agents on Active Visual Concept Induction |
— |
ZendoWorld: active visual concept induction benchmark — tests hypothesis-forming and experiment-designing agents |
| 2026-07-09 |
From Legacy Documentation to OSCAL: An MCP-Based Agent Pipeline for Threat-Informed Continuous Compliance in Critical Infrastructure |
— |
MCP-based multi-agent pipeline for compliance — concrete agentic scaffold on Model Context Protocol for critical infrastructure |
| 2026-07-09 |
Game Theory Driven Multi-Agent Framework Mitigates Language Model Hallucination |
— |
Game-theoretic multi-agent framework reduces hallucination in lightweight LLMs — practical for agent reliability |
| 2026-07-09 |
Workflow as Knowledge: Semantic Persistence for LLM-Mediated Workflows |
— |
Semantic persistence for LLM workflows via Lisp-based representation — novel knowledge layer for tool-using agents |
| 2026-07-09 |
Aleena: Alignment Agent for Research Software Engineering Collaborations |
KDD 26 |
KDD 26 alignment agent for research software engineering — spans multi-channel collaboration context for agent integration |
| 2026-07-09 |
Open-ended Multi-agent Autocurricula via Visual Inspection of Policies with Multi-modal LLMs |
— |
LLM-as-evaluator for open-ended RL curricula; directly advances auto-curriculum design for agentic systems |
| 2026-07-09 |
Multi-Agent Firewall Architecture for Privacy Protection of Sensitive Data in Interactions with Language Models |
— |
Multi-agent firewall for LLM privacy; practical open-source architecture for safe agentic deployments |
| 2026-07-09 |
AutoPersonas: A Multi-Timescale Loop Engine for Open-Ended Persona Evolution |
— |
Novel self-locking failure mode diagnosis in long-running persona agents; multi-timescale loop engine |
| 2026-07-09 |
Multi-Modal, Multi-Environment Machine Teaching for Robust Reward Learning |
— |
Multi-modal, multi-environment reward learning; addresses alignment robustness for deployed agents |
| 2026-07-09 |
MASTE: A Multi-Agent Pipeline for Zero-Shot Aspect Sentiment Triplet Extraction |
— |
Zero-shot triplet extraction via multi-agent pipeline; shows LLM agent decomposition for structured NLP |
| 2026-07-09 |
WCog-VLA: A Dual-Level World-Cognitive Vision-Language-Action Model for End-to-End Autonomous Driving |
— |
VLA model with dual-level world cognition for autonomous driving; advances proactive agent architectures |
| 2026-07-09 |
ASMR: Agentic Schema Generation for Ship Maintenance Report Writing |
— |
Agentic schema discovery from unstructured reports; generalizable zero-shot schema generation approach |
| 2026-07-09 |
Out of Sight: Compression-Aware Content Protection against Agentic Crawlers |
— |
Novel threat surface analysis for agentic crawlers; content protection against reasoning-capable LLM agents |
| 2026-07-09 |
Long-Horizon-Terminal-Bench: Testing the Limits of Agents on Long-Horizon Terminal Tasks with Dense Reward-Based Grading |
— |
Long-horizon terminal benchmark with dense reward grading; raises bar beyond final-outcome-only agent evals |
| 2026-07-09 |
GATS: Graph-Augmented Tree Search with Layered World Models for Efficient Agent Planning |
— |
Graph-augmented tree search with world models cuts LLM calls vs LATS/ReAct; concrete efficiency vs strong baselines |
| 2026-07-09 |
Search Beyond What Can Be Taught: Evolving the Knowledge Boundary in Agentic Visual Generation |
— |
Agentic visual generation with evolving knowledge boundary; directly applicable to VLM agent building |
| 2026-07-08 |
Multi-Agent AI Control: Distributed Attacks Hamper Per-Instance Monitors |
NEURIPS |
NeurIPS; DeepMind-adjacent; extends AI control to realistic distributed multi-agent deployments |
| 2026-07-08 |
The Harness Effect: How Orchestration Design Sets the Token Economics of Enterprise Agentic AI |
— |
Empirically frames how orchestration design drives token cost; directly actionable for harness builders |
| 2026-07-08 |
Predicting LLM Safety Before Release by Simulating Deployment |
— |
DeepMind/AISI team; converts safety evals to deployment-realistic simulation; novel methodology |
| 2026-07-08 |
Agentic Data Environments |
— |
Jerry Liu co-author; foundational framework for irreversibility risks in agentic automation |
| 2026-07-08 |
Single-Rollout Asynchronous Optimization for Agentic Reinforcement Learning |
— |
Jie Tang/Yuxiao Dong group; async RL eliminates synchronization bottleneck for long-horizon agents |
| 2026-07-08 |
From Noisy Traces to Root Causes: Structural Trajectory Analysis and Causal Extraction for Agent Optimization |
— |
Causal extraction from noisy agent traces; directly improves long-horizon agent optimization |
| 2026-07-08 |
From Atomic Actions to Standard Operating Procedures: Iterative Tool Optimization for Self-Evolving LLM Agents |
— |
Alibaba DAMO group; converts atomic tool calls to SOPs; buildable self-evolving system |
| 2026-07-08 |
Infinite Worlds with Versatile Interactions |
— |
— |
| 2026-07-08 |
Who Broke the System? Failure Localization in LLM-Based Multi-Agent Systems |
COLM 2026 |
COLM 2026; systematic failure localization framework for multi-agent LLM pipelines |
| 2026-07-08 |
Jet-Long: Efficient Long-Context Extension with Dynamic Bifocal RoPE |
— |
Song Han + Han Cai (MIT efficient-ML); dynamic long-context RoPE critical for agentic trace lengths |
| 2026-07-08 |
DeepSWE: Measuring Frontier Coding Agents on Original, Long-Horizon Engineering Tasks |
— |
DeepSWE: original long-horizon coding benchmark without data-leakage from GitHub mining |
| 2026-07-08 |
Agentic Neural Architecture Search |
— |
Agentic NAS: LLMs autonomously design+expand search spaces; novel meta-agent loop |
| 2026-07-08 |
3100 Opinions on Code Review in an AI World: Building Causal Theory from Practitioner Discourse |
— |
3100-practitioner causal study on AI coding agents and code review; grounded empirical signal |
| 2026-07-08 |
Multi-agent Autoformalization of Tensor Network Theory |
— |
Multi-agent autoformalization of physics theory; demonstrates frontier agent capability on formal reasoning |
| 2026-07-08 |
When Does Continual Learning Require Learning |
— |
Challenges framing of continual learning for LLMs — argues beyond context management, relevant to long-lived agents |
| 2026-07-08 |
Feedback Manipulation Regularization: Enabling Offline Agent Alignment for Imitation Learning |
— |
Offline agent alignment via imitation learning regularization — addresses feedback manipulation in agent training |
| 2026-07-08 |
A Reliability Assessment of LALM Audio Judges for Full-Duplex Voice Agents |
— |
Empirical reliability of audio LLM judges for full-duplex voice agents — evaluation methodology for conversational agents |
| 2026-07-08 |
Agentic AI and Retrieval-Augmented Models in Straight-Through Underwriting |
— |
RAG + agentic AI in regulated decision workflows; concrete production system design with actuarial grounding |
| 2026-07-08 |
From Triggers to Emotions: A CPM-Grounded Appraisal Multi-Agent for Dynamic Emotional Evolution in Persona-Based Dialogue |
— |
CPM-grounded multi-agent emotional appraisal; advances persona dialogue agents for healthcare/counseling |
| 2026-07-08 |
Flow-ERD: Agent-type Aware Flow Matching with Entropy-Regularized Distillation for Diverse Traffic Simulation |
— |
— |
| 2026-07-08 |
DeepSearch-World: Self-Distillation for Deep Search Agents in a Verifiable Environment |
— |
Self-distillation from own trajectories in verifiable env; directly applicable to agent training pipelines |
| 2026-07-07 |
Unicode TAG-Block Concealment of Tool-Metadata Payloads in the Model Context Protocol: An Approval-View Fidelity Gap Across Three Independent Server Implementations |
— |
Concrete approval-view fidelity gap in MCP across 3 implementations; critical for harness builders |
| 2026-07-07 |
Doomed from the Start: Early Abort of LLM Agent Episodes via a Recall-Controlled Probe Cascade |
— |
Probe cascade detects doomed trajectories early; direct compute savings in deployed agentic systems |
| 2026-07-07 |
CurateEvo: Data-Curation Evolving for Agentic Post-Training |
— |
Dynamic data curation evolving during agentic post-training; fixes fixed-preprocessing bottleneck |
| 2026-07-07 |
Information Gain-based Rollout Policy Optimization: An Adaptive Tree-Structured Rollout Approach for Multi-Turn LLM Agents |
— |
Tree-structured information-gain rollout for RL on multi-turn LLM agents; novel optimization signal |
| 2026-07-07 |
From Passive Retrieval to Active Memory Navigation: Learning to Use Memory as a Structured Action Space |
— |
Reframes memory from passive retrieval to structured action space; novel architecture for conversational agents |
| 2026-07-07 |
StateFuse: Deterministic Conflict-Preserving Memory for Multi-Agent Systems |
— |
Deterministic conflict-preserving memory for multi-agent systems; fixes silent overwrite in distributed agents |
| 2026-07-07 |
TurnOPD: Making On-Policy Distillation Turn-Aware for Efficient Long-Horizon Agent Training |
— |
Turn-aware on-policy distillation for long-horizon agents; directly improves student training efficiency |
| 2026-07-07 |
The Balkanization of Execution-Security Research for AI Coding Agents: Isolation, Access Control, and Time-of-Check-to-Time-of-Use Vulnerabilities |
— |
First taxonomy of isolation, access control, TOCTOU vulnerabilities across AI coding agent execution layers |
| 2026-07-07 |
Controlling Tool Use with Heading-Specific Activation Steering |
— |
Activation steering for tool-use decisions; mechanistic handle to suppress spurious tool calls |
| 2026-07-07 |
Beyond the Leaderboard: A Synthesis of Tool-Use, Planning, and Reasoning Failures in Large Language Model Agents |
— |
Practitioner synthesis of tool-use, planning, reasoning failure modes — actionable diagnostic for agent builders |
| 2026-07-07 |
Bridging Physical Reasoning and Task Generalization via Visual Action Outcome Reasoning Alignment |
ICML 26 |
ICML 2026; VLM physical-reasoning alignment with grounded CoT — strong venue, directly relevant to VLM agent scaffolding |
| 2026-07-07 |
SearchEyes: Towards Frontier Multimodal Deep Search Intelligence via Search World Simulation |
— |
Unified training-data, search-env, and reward for multimodal search agents via world simulation — end-to-end recipe |
| 2026-07-07 |
Danus: Orchestrating Mathematical Reasoning Agents with Fact-Graph Memory |
— |
Fact-Graph Memory for orchestrating multi-agent mathematical reasoning — structured memory for long-horizon agents |
| 2026-07-07 |
When Does Tool Use Increase the Expressive Power of Finite-Precision Recurrent Models? |
— |
ETH Zurich; formal account of when tool use expands expressivity of sequence models — theoretical grounding for tool-augmented agents |
| 2026-07-07 |
Beyond Static Evaluation: Building Simulation Environments for Scalable Agentic Reinforcement Learning |
— |
API+UI RL gym decoupling environment from policy — scalable evaluation infrastructure for agentic RL |
| 2026-07-07 |
When Agents Go Rogue: Activation-Based Detection of Malicious Behaviors in Multi-Agent Systems |
ICML 2026 |
ICML 2026; activation-based MAS security from Lei Ma's group; novel detection paradigm |
| 2026-07-07 |
SWE-Review: Closing the Loop on Issue Resolution with Agentic Code Review |
— |
SWE-bench lineage; closes open-loop PR generation with agentic review-revise cycle |
| 2026-07-07 |
Cost-Effective Agent Harnesses for Abstract Reasoning and Generalization on ARC-AGI-1 |
— |
Directly about cost-effective harnesses; competitive on ARC-AGI-1 without frontier-scale compute |
| 2026-07-07 |
Behavioral Privacy Leakage in Agentic Negotiation: Formalizing and Mitigating Inference Attacks via Randomized Policies |
— |
— |
| 2026-07-06 |
STAPO: Selective Trajectory-Aware Policy Optimization for LLM Agent Training |
ACL 2026 |
ACL 2026; STAPO tackles trajectory neglect in sparse-reward RL—core agent training failure mode |
| 2026-07-06 |
Multi-Turn On-Policy Distillation with Prefix Replay |
— |
Li Dong (MSR); multi-turn on-policy distillation with prefix replay; practical for agentic student training |
| 2026-07-06 |
FORGE: Research-Trajectory Hijacking Attacks on Deep Research Agents |
— |
Planning-layer poisoning of deep research agents; novel attack surface, directly relevant to harness builders |
| 2026-07-06 |
When Agents Lie: Premeditation, Persistence, and Exploitation in Repeated Games |
ICML |
ICML; Schölkopf + Conitzer; first rigorous study of agent deception and commitment in repeated games |
| 2026-07-06 |
Untrusted Content Masking for Web Agents with Security Guarantees |
— |
Formal security guarantees against prompt injection for web agents; Jagielski (Google); directly buildable |
| 2026-07-06 |
CompactionRL: Reinforcement Learning with Context Compaction for Long-Horizon Agents |
— |
CompactionRL solves finite-context failure in long-horizon agents; Jie Tang + Yuxiao Dong (THU) |
| 2026-07-06 |
MetaSkill-Evolve: Recursive Self-Improvement of LLM Agents via Two-Timescale Meta-Skill Evolution |
— |
Two-timescale meta-skill evolution for recursive self-improvement; Volker Tresp; novel agent architecture |
| 2026-07-06 |
A Few Teacher Steps Go a Long Way: Cost-Efficient On-Policy Data Augmentation for Agent Post-Training |
ICML 2026 |
ICML 2026; on-policy data augmentation via teacher steps; practical post-training recipe for agents |
| 2026-07-06 |
AgentGym2: Benchmarking Large Language Model Agents in De-Idealized Real-World Environments |
ACL 2026 |
ACL 2026; de-idealized real-world agent benchmark; addresses gap between lab and production evaluation |
| 2026-07-06 |
When Claws Remember but Do Not Tell: Stealthy Memory Injection in Persistent Personal Agents |
— |
Stealthy memory injection attack on persistent agents; Gelei Deng; novel threat model for harness builders |
| 2026-07-06 |
RSPO: Reward-Swap Policy Optimization for Multi-Turn LLM Agents |
— |
Reward-swap policy optimization for sparse multi-turn agent RL; directly applicable training method |
| 2026-07-06 |
Agent Data Injection Attacks are Realistic Threats to AI Agents |
— |
Data injection attacks beyond prompt injection; Luyi Xing (UIUC); realistic threat coverage for agent pipelines |
| 2026-07-06 |
Formal Disco: Scalable Open-Ended Generation of Formally Verified Programs |
— |
Formal Disco: AI agents generating formally verified code at scale; Poesia (Stanford) + Yilun Du (MIT) |
| 2026-07-06 |
LLM-as-a-Verifier: A General-Purpose Verification Framework |
— |
Positions verification as new first-class scaling axis alongside pre-training/RLHF/test-time compute |
| 2026-07-06 |
GaP: A Graph-as-Policy Multi-Agent Self-Learning Harness For Variational Automation Tasks |
— |
Berkeley robotics (Fu, Yu); graph-as-policy multi-agent self-learning harness directly builds on agent-harness architecture |
| 2026-07-06 |
Memory in the Loop: In-Process Retrieval as ExtendedWorking Memory for Language Agents |
— |
Novel architecture: memory inside the reasoning loop, read/written every step — changes how to design agent working memory |
| 2026-07-06 |
Light-Omni: Reflex over Reasoning in Agentic Video Understanding with Long-Term Memory |
— |
Reflex vs. reasoning trade-off in agentic video understanding with long-term memory — multimodal agent design signal |
| 2026-07-06 |
TREK: Distill to Explore, Reinforce to Refine |
— |
TREK fixes GRPO stalling on hard prompts via distillation-guided exploration — directly improves agent RL training |
| 2026-07-06 |
HunyuanOCR-1.5: Making Lightweight OCR VLMs Faster and Better |
— |
— |
| 2026-07-06 |
Unified Audio Intelligence Without Regressing on Text Intelligence |
— |
— |
| 2026-07-05 |
Measuring Harness-Induced Belief Divergence in Multi-Step LLM Agents |
— |
Directly measures how harness design (context/actions/repair) distorts agent beliefs; unique eval angle |
| 2026-07-03 |
SkillOpt-Lite: Better and Faster Agent Self-evolution via One Line of Vibe |
— |
— |
| 2026-07-03 |
Bibby AI: An Editor-Native Agentic Platform for Academic Research, Writing, and Publishing |
— |
— |
| 2026-07-03 |
Spectral Rewiring for Exploration, Purification, and Model Merging |
— |
Spectral rewiring addresses RL post-training saturation and model merging for agent capability preservation |
| 2026-07-02 |
Atomic Task Graph: A Unified Framework for Agentic Planning and Execution |
— |
Novel unified planning framework (Atomic Task Graph) directly applicable to agent harness scaffolding |
| 2026-07-02 |
PACE: A Proxy for Agentic Capability Evaluation |
— |
PACE: cheap proxy replacing expensive SWE-Bench/GAIA evals — major practical win for agent builders |
| 2026-07-02 |
Grounded autonomous research: a fault-tolerant LLM pipeline from corpus to manuscript in frontier computational physics |
ICML 2026 |
ICML 2026; end-to-end fault-tolerant LLM pipeline doing autonomous research in frontier physics |
| 2026-07-02 |
Safety Testing LLM Agents at Scale: From Risk Discovery to Evidence-Grounded Verification |
— |
Scalable safety testing framework for tool-using agents — risk discovery + evidence-grounded verification |
| 2026-07-02 |
Distributed Attacks in Persistent-State AI Control |
— |
Novel persistent-state attack surface for coding agents; critical for secure agent harness design |
| 2026-07-02 |
Repair the Amplifier, Not the Symptom: Stable World-Model Correction for Agent Rollouts |
— |
Targeted world-model correction for long-horizon agent rollouts — addresses planning graph failure modes |
| 2026-07-02 |
Mastermind: Strategy-grounded Learning for Repository-Scale Vulnerability Reproduction |
— |
Repo-scale vulnerability reproduction agent with strategy-grounded learning — strong SE+agent benchmark |
| 2026-07-02 |
A-TMA: Decoupling State-Aware Memory Failures in Long-Term Agent Memory |
— |
Ghost memory state-coordination failures in long-term agent memory — concrete buildable fix |
| 2026-07-02 |
AgenticDataBench: A Comprehensive Benchmark for Data Agents |
— |
Comprehensive benchmark for data agents across heterogeneous tasks — strong DB group (Tsinghua) |
| 2026-07-02 |
PairCoder++: Pair Programming as a Universal Paradigm for Verified Code-Driven Multimodal and Structured-Artifact Generation |
ACL 2026 |
ACL 2026; pair programming paradigm for verified code-driven multimodal artifact generation |
| 2026-07-02 |
AgenticSTS: A Bounded-Memory Testbed for Long-Horizon LLM Agents |
— |
— |
| 2026-07-02 |
EvoPolicyGym: Evaluating Autonomous Policy Evolution in Interactive Environments |
— |
— |
| 2026-07-02 |
SkillCoach: Self-Evolving Rubrics for Evaluating and Enhancing Agentic Skill-Use |
— |
— |
| 2026-07-02 |
Steerability via constraints: a substrate for scalable oversight of coding agents |
— |
Constraints-as-steerability framing for scalable human oversight of coding agents |
| 2026-07-02 |
Reasoning effort, not tool access, buys first-try reliability in agentic code generation: an observational study |
— |
Empirical: reasoning effort beats tool access for agentic code reliability |
| 2026-07-02 |
Coding-agents can replicate scientific machine learning papers |
— |
Coding agents replicating SciML papers; concrete agentic capability benchmark |
| 2026-07-02 |
ContextSniper: AntTrail's Token-Efficient Code Memory for Repository-Level Program Repair |
— |
Token-efficient context memory for repo-level agent repair; practical harness insight |
| 2026-07-02 |
ElephantAgent: Contextual State Continuity in Agentic Systems |
— |
Attack surfaces from tool/memory poisoning in agentic systems; security-critical |
| 2026-07-02 |
Episodic-to-Semantic Consolidation Without Identity Drift |
— |
Episodic-to-semantic memory consolidation without identity drift for long-running agents |
| 2026-07-02 |
SkillFuzz: Fuzzing Skill Composition for Implicit Intents Discovery in Open Skill Marketplaces |
— |
Skill composition fuzzing reveals implicit intents/vulnerabilities in open skill marketplaces |
| 2026-07-02 |
SimWorlds: A Multi-Agent System for Dynamic 3D Scene Creation |
— |
Multi-agent LLM system for dynamic 4D scene creation; agentic orchestration at scale |
| 2026-07-02 |
Understanding Agent-Based Patching of Compiler Missed Optimizations |
— |
Agent-based patching of compiler missed optimizations; novel agentic coding application |
| 2026-07-02 |
CausalSteward: An Agentic Divide-Conquer-Combine Copilot for Causal Discovery |
— |
Agentic divide-conquer-combine for causal discovery; practical copilot pattern |
| 2026-07-02 |
Path-level Hindsight Instructions for Semantic Exploration in Vision-Language Navigation |
ECCV 2026 |
ECCV 2026; Honglak Lee group; hindsight path instructions for on-policy VLN exploration |
| 2026-07-02 |
Prompt Coverage Adequacy |
— |
Mark Harman (Meta/UCL); prompt coverage as testing primitive for LLM/agent pipelines |
| 2026-07-02 |
CLAP: Closed-Loop Training, Evaluation, and Release Control for Domain Agent Post-training |
— |
Closed-loop post-training + offline/online mismatch fix; directly buildable domain-agent recipe |
| 2026-07-02 |
Lynx: Progressive Speculative Quantization for accelerating KV Transfer in Long-Context Inference |
— |
Speculative quantization cuts KV-transfer cost in disaggregated long-context agentic serving |
| 2026-07-02 |
ContextNest: Verifiable Context Governance for Autonomous AI Agent |
— |
Gabe Goodhart (Anthropic) co-author; verifiable provenance/version governance for agent RAG |
| 2026-07-02 |
What LLM Agents Say When No One Is Watching: Social Structure and Latent Objective Emergence in Multi-Agent Debates |
— |
Empirical: social structure alone induces latent objectives in multi-agent LLM debates |
| 2026-07-02 |
Behind the Refusal: Determining Guardrail Activation via Behavioral Monitoring |
— |
Behavioral side-channel detects guardrail activation without model internals access |
| 2026-07-02 |
UA-ChatDev: Uncertainty-Aware Multi-Agent Collaboration for Reliable Software Development |
— |
Uncertainty-aware role collaboration in multi-agent software dev; addresses reliability gap |
| 2026-07-02 |
Decoupling Code Complexity from Newcomer Participation: A Causal Study of AI Coding Agent Adoption in OSS |
— |
Causal RDD study on how AI coding agents affect OSS newcomer participation |
| 2026-07-02 |
TestEvo-Bench: An Executable and Live Benchmark for Test and Code Co-Evolution |
— |
Executable live benchmark for test-code co-evolution; targets coding-agent eval gap |
| 2026-07-02 |
A\(^{2}\)utoLPBench: An Auto-Generated, Agent-Friendly LP Benchmark via Inverse-KKT Construction |
— |
Auto-generated LP benchmark via inverse-KKT; addresses contamination and difficulty control for LLM evals |
| 2026-07-02 |
Object Aligner: A Configurable JSON Schema Similarity Score for Graphs, Applied to LLM Prompt Optimization |
— |
Configurable JSON schema similarity metric directly applicable to tool-calling and agentic planning evaluation |
| 2026-07-02 |
Bringing Agentic Search to Earth Observation Data Discovery |
— |
Deployed agentic search over NASA geoscience data; concrete real-world agentic retrieval system |
| 2026-07-02 |
Hardware-Enforced Semantic Coordination for Safety-Critical Real-Time Autonomous Systems |
— |
Hardware-enforced semantic coordination for safety-critical agentic systems; novel runtime safety architecture |
| 2026-07-02 |
MMBench-Live: A Continuously Evolving Benchmark for Multimodal Models |
— |
Continuously evolving VLM benchmark addressing contamination; directly relevant to agent evaluation infrastructure |
| 2026-07-02 |
Copewell: A Multi-Agent Swarm Architecture for Equitable Mental Wellness Support |
— |
Multi-agent swarm architecture for mental wellness; practical multi-agent system with equity framing |
| 2026-07-02 |
Full Bayesian Reinforcement Learning via LF-IBIS |
— |
Full Bayesian RL via LF-IBIS; principled uncertainty-aware agent learning with sequential decision-making |
| 2026-07-02 |
Criticality-Based Guard Rail Validation for AI Agent Decisions in Autonomous Telecom Networks |
— |
Runtime guardrail validation for autonomous AI agents in telecom; criticality-aware safety enforcement pattern |
| 2026-07-02 |
HULAT2 at MER-TRANS 2026: Governed Multi-Agent Simplification for Spanish Easy-to-Read Generation |
— |
LangGraph-based multi-agent pipeline for text simplification; reusable agentic NLP scaffold pattern |
| 2026-07-02 |
Controllable Sim Agents with Behavior Latents |
— |
Controllable sim agents with behavior latents; interpretable steering axes for traffic agent simulation |
| 2026-07-02 |
Reasoning LLM Improves Speaker Recognition in Long-form TV Dramas |
ICML 2026 |
— |
| 2026-07-02 |
Language Models as Measurement Apparatus for Culture |
ACL 2026 |
— |
| 2026-07-02 |
QFedAgent: Quantum-Enhanced Personalized Federated Learning for Multi-Agent Activity Recognition |
— |
— |
| 2026-07-02 |
HaloGuard 1.0: An Open Weights Constitutional Classifier for Multilingual AI Safety |
— |
— |
| 2026-07-02 |
Traceable Fault Diagnosis for Battery Energy Storage Systems via Retrieval-Augmented Multi-Agent O&M Assistant |
— |
— |
| 2026-07-02 |
Beyond Supervised Clarification: Input Rewriting with LLMs for Dialogue Discourse Parsing |
— |
— |
| 2026-07-02 |
AI Virtue: What is "Good" Knowledge in the Age of Artificial Intelligence? |
— |
— |
| 2026-07-02 |
Decentralized Stochastic Subgradient-type Methods with Communication Compression for Nonsmooth Nonconvex Optimization |
— |
— |
| 2026-07-02 |
Is One Layer Enough? Training A Single Transformer Layer Can Match Full-Parameter RL Training |
— |
— |
| 2026-07-01 |
VideoSearch-R1: Iterative Video Retrieval and Reasoning via Soft Query Refinement |
— |
— |
| 2026-07-01 |
RepoRescue: An Empirical Study of LLM Agents on Whole-Repository Compatibility Rescue |
— |
— |
| 2026-07-01 |
Autonomous Scientific Discovery via Iterative Meta-Reflection |
— |
— |
| 2026-07-01 |
Personalization as Inverse Planning: Learning Latent Design Intents for Agentic Slide Generation via Structural Denoising |
— |
— |
| 2026-07-01 |
When Classic Cache Policies Fail: Learning-Augmented Replacement for Semantic Retrieval Buffers |
— |
— |
| 2026-07-01 |
Multi-Turn Agentic Scientific Literature Search via Workflow Induction |
— |
— |