> The idea of someone carrying out a crime via a proxy is not new; the main difference now is that the proxy may be an AI. I thought that the concept discussed

> The idea of someone carrying out a crime via a proxy is not new; the main difference now is that the proxy may be an AI. I thought that the concept discussed in the OP was slightly different (and possibly even more insidious) : The proxy is a human who carries out orders of (or covers for) an AI that is known to be unsafe. I.e. an autonomous car manufacturer may acknowledge that their AI sometimes makes errors which can lead to accidents - so they equip their cars with a steering wheel and mandate that a human driver must always be present who can takeover from the AI in dangerous situations. The problem is just that the human driver may not realistically be able to do that, because reaction times are too short, the AI releases control too late, etc. Yet, if the AI malfunctions and the driver doesn't manage to prevent the accident, the company can comfortably blame the driver and keep the AI out of the discussion.