412 AI skill assessments later — the one metric nobody expects

We've been running aisa.to for a while now — conversational AI skill assessments, not quizzes. 412 completed.

The number that keeps surprising people: Safety scores average 45/100. Lowest dimension by a wide margin.

Not prompting. Not tool knowledge. Safety. As in — knowing when NOT to trust AI output, understanding data privacy implications, having an actual process for verifying what the model gives you.

62% of people score below Proficient overall (aisa.to/state-of-ai-fluency). But on Safety specifically, even people who nail everything else tend to fall short. The gap between "I use AI daily" and "I use AI responsibly" is real.

This week we also shipped AISApedia — a searchable encyclopedia of 147 AI terms (aisa.to/resources). Came directly from assessment conversations. People kept asking "wait, what's RAG?" mid-assessment so we built the reference they needed.

Next up: tracking whether Safety scores improve on retakes. Early signal says yes but the sample is still small.