safety scored lowest in our AI assessment data — here's what that means for product teams

we've assessed 2,000+ professionals' AI fluency at aisa.to and the data keeps telling the same story: Safety scores average 45/100, the lowest of any dimension we measure.

what's surprising isn't that people don't know about AI safety. most do, at least conceptually. the gap is between knowing you should verify outputs and actually having a process to do it.

the pattern we see constantly: someone uses AI daily, considers themselves proficient, but when we dig into their actual verification habits it's basically "does this sound right?" that's not verification, that's vibes.

three things we changed in the product based on this:

1. we now surface verification habits explicitly in the assessment conversation
2. we added a safety-specific section to every report with concrete recommendations
3. we started framing safety as a skill gap, not a knowledge gap — it lands better with teams

the takeaway for other founders building AI tools: your users probably know they should be careful with AI output. they just don't have a system for it. build the system into your product instead of hoping they figure it out.

full data: aisa.to/state-of-ai-fluency