Comment by cobbzilla

As I understood it, the AI company is trying to prevent itself and its customers from engaging in (probably unintentional, but any) criminal activity.

When the AI company is "policing" your agent by requiring a "human in the loop", it's just CYA (cover your ass) for the AI company.

If your agent goes off and does something illegal, the AI company would be liable unless they have some legal deniability. By requiring you the human account owner to sign-off on what your agent is doing, you become liable for any crimes your agent commits on your behalf. I haven't read their TOS but I can guarantee there is some clause like this in there.

You are still completely free to commit crimes with your agent and suffer whatever legal consequences follow!! You just have to be clear that you intentionally wanted those actions to occur which resulted in the crime. If you repeatedly allow your agents to take actions that could potentially be criminal without any human-in-the-loop, they're going to ban you because it exposes themselves to potential criminal charges.