What is something current AI systems are very good at, but people still don’t trust them to do?

We see benchmarks and demos showing strong performance, but hesitation still shows up in real use. Curious where people draw the trust line and why, whether it’s technical limits, incentives, or just human psychology.