Consistency under paraphrase in medical VLMs is a false proxy for reliability that hides models ignoring visual inputs entirely.
March 24, 2026
Original Paper
Consistent but Dangerous: Per-Sample Safety Classification Reveals False Reliability in Medical Vision-Language Models
arXiv · 2603.20985
The Takeaway
It reveals a 'Dangerous' quadrant where models achieve 99% accuracy and perfect consistency by relying solely on text patterns, making them invisible to standard confidence-based screening. This mandates a change in how medical vision-language models are evaluated, requiring a text-only baseline to detect image-independent predictions.
From the abstract
Consistency under paraphrase, the property that semantically equivalent prompts yield identical predictions, is increasingly used as a proxy for reliability when deploying medical vision-language models (VLMs). We show this proxy is fundamentally flawed: a model can achieve perfect consistency by relying on text patterns rather than the input image. We introduce a four-quadrant per-sample safety taxonomy that jointly evaluates consistency (stable predictions across paraphrased prompts) and image