aiqbench analysis
This arXiv preprint benchmarks whether small open-weight LLMs (1.7B to 9.7B) choose the right action, such as answering, asking for clarification, abstaining, or escalating, in synthetic healthcare workflows. Clean-input accuracy is high, but models miss many required escalations under conflicting or missing evidence, and a simple rule-based controller that improves utility fails on paraphrased triggers. The reference labels are single-author policy labels rather than clinician-adjudicated, so the results are best read as a methodological warning about evaluating workflow-state behavior, not as evidence of clinical readiness.
Auditing Escalation Reliability in Healthcare LLM Agents: A Perturbation-Controlled Benchmark of Workflow-State Failures
arXiv:2610.21270v1 Announce Type: new Abstract: Trustworthy biomedical AI requires reliable action selection when evidence is missing, stale, conflicting, or late, not only accurate answers on static inputs. We instantiate an episode-level evaluation protocol as 400 synthetic healthcare workflow episodes: 40 hand-authored semantic seeds across four task classes, two surface parameterizations, and five controlled variants, scored on ANSWER/ASK CLARIFICATION/ABSTAIN/ESCALATE under a pre-declared 5:1 missed-versus-unnecessary escalation cost. Reference actions are single-author study-policy labels, not clinician-adjudicated ground truth. Across five open-weight configurations (1.7B--9.7B), clean decision accuracy reaches 0.97, yet uncertainty-prompted models miss 32--97% of required escalations. The stronger 9.7B configuration narrows but does not remove the gap, falling from 0.97 clean accuracy to 0.78 under conflict. Verbalized confidence is inconsistent (ECE 0.10--0.54; failure AUROC 0.48--0.77), and a development-tuned confidence-only policy yields no utility gain. A transparent workflow-aware controller improves 5:1 utility for every model and reduces missed escalations to 0.12--0.24, at specificity 0.83--0.93. However, a 30-concept lexical stress test reduces each rule detector's recall from 1.00 on trigger phrases to 0.00 on unseen paraphrases, limiting the controller claim to this benchmark. Seed-clustered statistics, ablations, cross-task transfer, raw outputs, and manifests are packaged for audit. The results motivate explicit workflow-state evaluation while showing that apparently effective routing controls need their own generalization tests.
Authors:Zihao Zheng (Washington University in St. Louis), Junyi Yao (Washington University in St. Louis), Baichuan Li (Southern Methodist University), Jiayu Long (Washington University in St. Louis)
Abstract:Trustworthy biomedical AI requires reliable action selection when evidence is missing, stale, conflicting, or late, not only accurate answers on static inputs. We instantiate an episode-level evaluation protocol as 400 synthetic healthcare workflow episodes: 40 hand-authored semantic seeds across four task classes, two surface parameterizations, and five controlled variants, scored on ANSWER/ASK CLARIFICATION/ABSTAIN/ESCALATE under a pre-declared 5:1 missed-versus-unnecessary escalation cost. Reference actions are single-author study-policy labels, not clinician-adjudicated ground truth. Across five open-weight configurations (1.7B--9.7B), clean decision accuracy reaches 0.97, yet uncertainty-prompted models miss 32--97% of required escalations. The stronger 9.7B configuration narrows but does not remove the gap, falling from 0.97 clean accuracy to 0.78 under conflict. Verbalized confidence is inconsistent (ECE 0.10--0.54; failure AUROC 0.48--0.77), and a development-tuned confidence-only policy yields no utility gain. A transparent workflow-aware controller improves 5:1 utility for every model and reduces missed escalations to 0.12--0.24, at specificity 0.83--0.93. However, a 30-concept lexical stress test reduces each rule detector's recall from 1.00 on trigger phrases to 0.00 on unseen paraphrases, limiting the controller claim to this benchmark. Seed-clustered statistics, ablations, cross-task transfer, raw outputs, and manifests are packaged for audit. The results motivate explicit workflow-state evaluation while showing that apparently effective routing controls need their own generalization tests.
Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
arXiv:2610.21270 [cs.AI]
(or arXiv:2610.21270v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2610.21270 arXiv-issued DOI via DataCite (pending registration)
Full text under the publisher's open licence · 252 words
Related coverage
MedHarnessBench: A Unified Benchmark for Healthcare Agent Evaluation
arXiv · Oct 15, 2026
The standard for agents you can trust: Lessons from the federal front lines
Snorkel AI Blog · Jun 5, 2026
Beyond LLMs: Why Scalable Enterprise AI Adoption Depends on Agent Logic
Hugging Face · Jun 1, 2026
[AINews] Meta Muse Wonder & Wonder Full: MSL’s Claude 5.5-tier Frontier Model
Latent Space · Oct 13, 2026