Long ago, when I were immensely less rational, I actually strongly believed somehting very similar to this, and acted on this belief trying to stop my mind from creating models of people. I still feel uneasy about creating highly detailed characters. I probably would go "I knew it!" if the AI said this.

Upvoted for reminding me of 1/0 (read through 860).

I find the idea that they're conscious more likely than the idea that death is inherently bad. I also doubt that they're as conscious as humans (either it isn't discrete, and a human is more, or it is, and a human has more levels of consciousness), and that their emotions are what they appear to be.

Even more sinister, maybe: suppose it said there's a level of processing on which you automatically interpret things in an intentional frame (ala Dan Dennet) and this ability to "intentionalize" things effectively simulates suffering/minds all the time in everyday objects in your environment, and that further, while we can correct it in our minds, this anthropomorphic projection happens as a necessary product, somehow, of our consciousness. Consciousness as we know it IS suffering and to create an FAI that won't halt the moment it figures out that it is causing harm with its own thought processes, we'll need to think really, really far outside the box.