Debate: AI agents may pose less insider risk than human employees

quetzal_rainbow · x · 2026-10-21

quetzalrainbow argues near-term AI misalignment is overhyped: AI agents betray their users less often than human employees betray employers, so insider risk relative to economic value is lower. jessicata counters that data leakage is more an adversarial robustness issue than a motives issue, and that we have no empirical numbers on agent trustworthiness — quetzalrainbow admits there's no empirical test at hand either.

Original post →