Who&When Pro Benchmarks LLM Failure Attribution in AI Agents.

Jiale Liu, Huajun Xi, Shaokun Zhang, Yifan Zeng, Tianwei Yue, Chi Wang, Jian Kang, Qingyun Wu, Huazheng Wang· July 14, 2026 View original

Summary

This research introduces Who&When Pro, a large-scale benchmark for automated failure attribution in agentic AI systems, featuring 12,326 failed trajectories with golden labels across diverse modalities and scenarios. It reveals systematic patterns in how LLMs attribute failures, offering empirical guidance for future attribution systems.

As AI agents become increasingly sophisticated, identifying the root causes of their failures becomes more challenging yet crucial. This research addresses this by introducing "Who&When Pro," a comprehensive, large-scale benchmark specifically designed for automated failure attribution in agentic AI systems. The benchmark is meticulously constructed using a controlled pipeline that injects failures only after successfully replaying a prefix of a successful trajectory, ensuring precise ground truth. Who&When Pro comprises 12,326 failed trajectories, each with golden labels, spanning three different modalities and 26 distinct benchmarks. These cover a wide array of scenarios, providing a robust dataset for evaluating how well Large Language Models (LLMs) can pinpoint where and why an agentic system failed. Beyond just benchmarking, the study includes extensive experiments and analyses that uncover systematic patterns in how various LLM families attribute failures across different modalities and protocols. These findings offer valuable empirical guidance for the development and improvement of future automated failure attribution systems, which are essential for debugging, enhancing reliability, and advancing the capabilities of complex AI agents.

Why it matters

For professionals developing, deploying, or managing AI agents, accurate failure attribution is critical for debugging, improving system reliability, and accelerating development cycles. This benchmark provides tools and insights to achieve that.

How to implement this in your domain

  1. 1Utilize the Who&When Pro benchmark to evaluate the failure attribution capabilities of your current LLM-based agent systems.
  2. 2Analyze the systematic patterns identified in the research to inform the design of your own attribution mechanisms.
  3. 3Integrate automated failure attribution tools into your AI agent development and monitoring pipelines.
  4. 4Experiment with different LLM families and protocols for attribution to find the most effective approach for your specific agents.
  5. 5Contribute to the development of more robust failure attribution systems by leveraging insights from this benchmark.

Who benefits

AI/ML DevelopmentSoftware EngineeringRoboticsAutonomous SystemsQuality Assurance

Key takeaways

  • Who&When Pro is a new, large-scale benchmark for automated failure attribution in AI agents.
  • It provides 12,326 labeled failed trajectories across diverse scenarios.
  • The research reveals systematic patterns in LLM failure attribution.
  • This benchmark offers empirical guidance for developing better attribution systems.

Original post by Jiale Liu, Huajun Xi, Shaokun Zhang, Yifan Zeng, Tianwei Yue, Chi Wang, Jian Kang, Qingyun Wu, Huazheng Wang

"arXiv:2607.09996v1 Announce Type: new Abstract: Automated failure attribution uses LLMs to identify where and why agentic systems fail. As agents become more capable, their failures become subtler, making automated attribution increasingly important. We introduce Who&When Pro, a…"

View on X

Originally posted by Jiale Liu, Huajun Xi, Shaokun Zhang, Yifan Zeng, Tianwei Yue, Chi Wang, Jian Kang, Qingyun Wu, Huazheng Wang on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses