mlsanityData to 7 Oct 2026

Towards a Science of AI Agent Reliability

arXiv 2602.16666 ↗First posted 18 Feb 2026cs.AI cs.CY cs.LGcited by 58

Not on the map. Why?

Cited by 30 in LLM agents and tool use, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

AI agents are increasingly deployed to execute important tasks. While rising accuracy scores on standard benchmarks suggest rapid progress, many agents still continue to fail in practice. This discrepancy highlights a fundamental limitation of current evaluations: compressing agent behavior into a single success metric obscures critical operational flaws. Notably, it ignores whether agents behave consistently across runs, withstand perturbations, fail predictably, or have bounded error severity. Grounded in safety-critical engineering, we provide a holistic performance profile by proposing twelve concrete metrics that decompose agent reliability along four key dimensions: consistency, robustness, predictability, and safety. Evaluating 15 models across two complementary benchmarks, we find that recent capability gains have only yielded small improvements in reliability. By exposing these persistent limitations, our metrics complement traditional evaluations while offering tools for reasoning about how agents perform, degrade, and fail.

Key facts. Cited by 58 papers in the map (data through 7 Oct 2026). First posted on arXiv 18 Feb 2026.

Builds on

0 of 0 references.

No references here yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 58.

58 papers.

Sort:

By area:

  1. 2026Log analysis is necessary for credible evaluation of AI agentscited by 8
  2. 2026SafetyDrift: Predicting When AI Agents Cross the Line Before They Actually Docited by 4
  3. 2026Hallucinations Undermine Trust; Metacognition is a Way Forwardcited by 4
  4. 2026High-Risk AI Systems and the Problem of Identity in the European AI Actcited by 4
  5. 2026The Implementation Lottery: Auditing Idea Reliability in Automated Researchcited by 4
  6. 2026Consistency as a Testable Property: Statistical Methods to Evaluate AI Agent Reliabilitycited by 3
  7. 2026Open-World Evaluations for Measuring Frontier AI Capabilitiescited by 3
  8. 2026Life After Benchmark Saturation: A Case Study of CORE-Benchcited by 3
  9. 2026RemoteShield: Enable Robust Multimodal Large Language Models for Earth Observationcited by 2
  10. 2026Designing for Doubt: The Case for Informed Abstention in Autonomous Agentscited by 2
  11. 2026How Much Coordination Gain Is Real? A Paired Noise-Floor Protocol for Multi-Agent LLM Benchmarkscited by 2
  12. 2026Who Checks the Citations? Benchmarking Legal Hallucination Detectioncited by 2
38 more
  1. 2026A Category Theory Account of AI Identitycited by 2
  2. 2026Decision Making Needs Uncertainty Quantification [Lecture Notes]cited by 2
  3. 2026Plover: Steering GUI Agents through Plan-Centric Interactioncited by 2
  4. 2026The Devil Is in the Interface: Evaluating How Tool Architecture Shapes Coding Agent Behaviorcited by 2
  5. 2026On the Fragility of Self-Improving Agents: Variance, Task Order, and Underspecificationcited by 2
  6. 2026The Double Measurement Confound in Agent Benchmarks: De-Scaffolding, Ground-Truth Scoring, and Reliability Beyond the Meancited by 2
  7. 2026Generating Expressive and Customizable Evals for Timeseries Data Analysis Agents with AgentFuelcited by 1
  8. 2026MarketBench: Evaluating AI Agents as Market Participantscited by 1
  9. 2026Agent Meltdowns: The Road to Hell Is Paved with Helpful Agentscited by 1
  10. 2026Update Opacity: Epistemic Accessibility and Governance Under AI System Changecited by 1
  11. 2026The "I Don't Know" Filter: Enhancing Agentic Reliability in Function Callingcited by 1
  12. 2026The Illusion of Robustness: Aggregate Accuracy Hides Prediction Flips under Task-Irrelevant Contextcited by 1
  13. 2026A Methodology for Auditable Trustworthiness Levels in AI Lifecycle Governancecited by 1
  14. 2026Property-Level Reconstructability of Agent Decisions: An Anchor-Level Pilot Across Vendor SDK Adapter Regimescited by 0
  15. 2026PocketAgents: A Manifest-Driven Library of Autonomous Defense Agentscited by 0
  16. 2026VET: A Framework for Analyzing AI Discoursecited by 0
  17. 2026Monitoring Agentic Systems Before They're Reliablecited by 0
  18. 2026The Agentic Web Requires New Normative Infrastructurecited by 0
  19. 2026ClayBuddy: A Framework, Evaluation, & Mitigation of Coding Agent Failurescited by 0
  20. 2026Reliability Scales Inversely: Hallucinations Snowball Faster in Bigger Language Modelscited by 0
  21. 2026AI Evaluation Should Measure Verification Cost, Not Correctness Alonecited by 0
  22. 2026No Task Fails Every Time: Why One-Shot Audits Are Structurally Blind to Agent Damagecited by 0
  23. 2026Testing and Evaluation of Agentic AI Systems In Military Command and Controlcited by 0
  24. 2026Using profiles of cognitive capability to assess AI suitability for workplace taskscited by 0
  25. 2026Resource Constraints and Performance in Agentic AI Systemscited by 0
  26. 2026Forward-Deployed Full-Stack Engineering for Autonomous Cloud MLOpscited by 0
  27. 2026Closing the Consistency Gap: Self-Evolving Agents That Learn to Stay on Coursecited by 0
  28. 2026Defining AI Agents: A Compendium of Criteria, Metrics, and Benchmarkscited by 0
  29. 2026When Hiring Becomes Agent-Mediated: Evaluating Access and Recurrence in Two-Agent R'esum'e Screeningcited by 0
  30. 2026LEGIT: Credentialing Protocol for Trustworthy AI Agent Marketplacescited by 0
  31. 2026EnterpriseVal: Quantifying the Efficacy, Reliability and Value of Generative AI in the Enterprisecited by 0
  32. 2026Convex AI Compositionality and the Governance of AI System Populationscited by 0
  33. 2026Risk-Aware Online Conformal State Probingcited by 0
  34. 2026Verifiable Hidden Dynamics Play: Generating Agentic RL Environments from Solved Mechanismscited by 0
  35. 2026RECLAIM: Can Agents Reproduce the Claims of Machine Learning Papers?cited by 0
  36. 2026Where Does Exactly-Once Live? Model, Harness, and Tool-Contract Effects on Duplicate Side Effects in LLM Agentscited by 0
  37. 2026Identical Runs, Different Results: Benchmarking AI Coding Agents on Open-Weight Modelscited by 0
  38. 2026The Marathon of Scientific Reasoning: Robustness of Scientific Agents to Perturbations in Multi-Turn Interactionscited by 0

50 most cited of 58 All 58 (JSON)