AI models would rather guess than ask for help, researchers find
22 AI models tested. Almost none ask for help when they don't know. Here's why that matters for deployment.
Why it matters
Academic research reveals a critical behavioral gap in multimodal AI models—they hallucinate rather than seek clarification—with implications for real-world reliability and user trust in production systems.
The key facts
10 to knowProactiveBench benchmark tests whether multimodal models request help when visual information is missing
22 models tested; almost none exhibit help-seeking behavior
Models default to guessing/hallucination over asking clarifying questions
Reinforcement learning approach shown as potential remediation pathway
Addresses fundamental safety/reliability concern for deployment-ready AI systems
ProactiveBench benchmark tests 22 multimodal language models
Almost no models ask for help when visual information is missing
Models default to guessing/hallucination over clarification requests
Reinforcement learning approach shows potential fix
Published April 2026
Go to the source
The Decoderthe-decoder.com
Publisher excerpt: ProactiveBench tests whether multimodal language models ask users for help when visual information is missing. Out of 22 models tested, almost none ask for what they need, but a simple reinforcement learning approach hints at a fix.
