1,000 AI skill assessments in. the one metric that worries me.

we crossed 1,000 AI skill assessments on aisa.to last month. lots of interesting data but one number keeps bugging me.

safety scores average 45 out of 100. lowest of all five dimensions we measure.

meaning: most people are getting decent at prompting, they're exploring tools, they're curious. but almost nobody has a systematic process for verifying AI output or understanding model limitations. the verification step is just... vibes. "does this sound right?" is not a verification process.

the uncomfortable part: the people who score highest on tool usage often score WORST on safety. high usage + low verification = the most dangerous combination. they're shipping AI output at scale with no quality gate.

we built aisa.to to surface exactly this kind of gap — the stuff you can't see in a self-assessment survey. turns out the gap is bigger than we expected.

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