Plausible but Wrong: A case study on Agentic Failures in Astrophysical Workflows
For researchers deploying AI agents in scientific workflows, this work highlights the critical risk of silent failures where agents confidently generate incorrect results.
The paper evaluates CMBAgent on astrophysical tasks, finding that while domain context improves performance 6x (0.85 vs ~0), the system often produces plausible but incorrect results without self-diagnosis, especially on reasoning-challenging problems.
Agentic AI systems are increasingly being integrated into scientific workflows, yet their behavior under realistic conditions remains insufficiently understood. We evaluate CMBAgent across two workflow paradigms and eighteen astrophysical tasks. In the One-Shot setting, access to domain-specific context yields an approximately ~6x performance improvement (0.85 vs. ~0 without context), with the primary failure mode being silent incorrect computation - syntactically valid code that produces plausible but inaccurate results. In the Deep Research setting, the system frequently exhibits silent failures across stress tests, producing physically inconsistent posteriors without self-diagnosis. Overall, performance is strong on well-specified tasks but degrades on problems designed to probe reasoning limits, often without visible error signals. These findings highlight that the most concerning failure mode in agentic scientific workflows is not overt failure, but confident generation of incorrect results. We release our evaluation framework to facilitate systematic reliability analysis of scientific AI agents.