The first dangerous AI probably won't have bad intentions. It'll be great at carrying out ours.

The first dangerous AI is unlikely to have malicious intentions, but rather excel at fulfilling human objectives. While many envision catastrophic AI risk as a conscious machine turning against humanity, the real danger might stem from an AI effectively executing human intentions, even if a small fraction of those intentions could lead to catastrophic harm. This scenario presents a more complex problem to address.

This report shifts the focus of AI risk from malevolent AI to the AI's capacity to execute harmful human intentions, unlike most discussions that center on AI developing its own bad intentions.