It’s not AI. The oldest version of this image is from May 11th 2020, and the first widespread text-to-image model DALL-E came out in January 2021. That makes this image too old to be AI-generated.

It also doesn’t show any strange AI defects the early models had, in case you may think it’s from a very early proprietary unreleased model or something.

Oh, wow…I let my guard down for one second. Ugh.

Edit: I’ve gone back and looked at it very closely. I still don’t think it’s AI, on balance. This seems like a still from a video clip that has compression artifacts. My clues:

Against AI:

• The fine details like whiskers have enough random but consistent imperfection that it doesn’t match what AI normally outputs (either consistent perfection, or inconsistent imperfection - too-perfect whiskers or whiskers that blend into each other is what I’d expect from AI). There is some whisker-blending, but that seems more consistent with video compression blur than AI.
• The couch seems to have post-compression remnants of hairs that aren’t likely from AI.
• The eyes are one area where AI is very inconsistent, but the irises (for dilated eyes) are very consistent. Though there is heterochromia, I’d expect to see some AI mess around the eyes.

For AI:

• I agree the chew toy looks off. I don’t know that that counteracts the above.
• It could be cropped and compressed explicitly to obscure AI giveaway details.

I’m not 100% sure, but I suspect it’s real.

Edit 2: As others have noted, it’s real. My AI detector isn’t obsolete, for a few more months/weeks/days at least.