Seems like this can be simply solved by amending the 3 laws with a 4th. 4. An AI must identify itself as an AI when asked. With this, an AI can trust the resp

Seems like this can be simply solved by amending the 3 laws with a 4th. 4. An AI must identify itself as an AI when asked. With this, an AI can trust the response of 'human'. A response of 'AI' is either an AI or a human lying at their own peril.