It's always like this. "Our AI did a very scary thing during an internal and unverifiable test/simulation/roleplay that can't apply to the real world use of our products! Quick news media, look how powerful and scary our AI is!"