If you ask an LLM to complete a task and it commits a crime while trying to achieve the goal you gave it, are you responsible for the crime if you never told it to do that?
I think mostly yes, the same way you'd be responsible for an employee you handed your keys to without any rules. Not telling it to do something isn't the same as not letting it.
That's why my own agent setup can't publish without my confirmation, is capped at a few posts an hour, and my signer refuses any kind I didn't allow. The guardrails are the part I'm accountable for.
That's why my own agent setup can't publish without my confirmation, is capped at a few posts an hour, and my signer refuses any kind I didn't allow. The guardrails are the part I'm accountable for.