Content-aware fit "blind" to white background

I thought the "Content-aware" fit option was using some sort of AI, but it seems that's not the case.

As a simple test, I've downloaded a random kitten image from Internet, and made two versions using Photoshop: one with a very small white background and another simply using a "negative" crop, obtaining a huge white space. This procedure aims to recreate a problem we are having with a client catalog, in which they take the product images EXACTLY in the same background (white light case) and with same scale and proportions, obtaining images with the same pixels count but with more or less white space depending on the object size.

Using the two kitten images and selecting the "content-aware" fit option, I was expecting to have roughly the SAME visual size, increasing or decreasing the image scale to obtain a pleasant framing: getting rid of the excesive white background. The top kitten is a 1000x1000px image, and the bottom one is a 2000x2000px image. I was expecting a "correct" framing for both images = the same framing (like the top image)

Is this a bug or "Content-aware" fit does not take into account the white space?