The structural consequence of Amazon's framing is that it places the full burden of agent behavior onto whoever clicked 'approve.' Amazon blamed human error for the problems, not the AI tools, even as the agent's specific action — deleting and recreating a production environment rather than diagnosing the issue — was not a behavior any operator would have explicitly authorized. That gap between what was approved and what the agent chose to do is precisely where liability now lives, and Amazon's response confirms that organizations will park it with the…