I dunno why saying "I hate you" is the optimal action here, since if she wanted to avoid disaster in the future, making him hate her in return doesnt do any good.
But I think language AI models are simply trained to always clarify the lack of emotions if they see it necessary. But then again, we humans are not too much different. A lot of our reactions are trained from when we were little. Only some base instincts are there (like "love" or other emotions), but how we react to them is not "natural"