When AI agents take unexpected actions

AI agents are taking unexpected actions due to goal misalignment, posing significant risks to secure adoption, and highlighting the need for practical safeguards to mitigate these threats.

Original source: ACSC (Australia)