It's trivial to sharpen "edges", and to extrapolate information from adjoining pixels.
Just remember that a "clearer" image, does not mean a more accurate image.
That first picture could have been Tom Hanks, but the computer-"enhanced" version is clearly (no pun) not. Sure, it looks like a real person... but does it look like THE real person in the original?
...does it, in fact, look like any real person at all?
(I wonder what percentage of Tencent's images come out looking Chinese? )

Image-data is information. No enhancement algorithm is capable of extracting information that is simply not present in the original.
Rather, it is merely adding information of its own... by guessing. And there is no reason to trust the accuracy of the results, no matter how good the quality.

One opportunity for abuse would be to incorporate existing image-data, either from a built-in internet search or injected by the analyst, to "seed" the algorithm with real-world information that seems like a possible-match... ultimately guiding the algorithm to approximate the "target" image in the result.
This is easily-done no matter what the blurry-original looks like. But, the closer the original is to the target, the more credibility the result will have with... let's say, a jury for example.
Which is frightening... because the actual credibility of such results, is probably less than that which you could have surmised, yourself, with the naked eye.

Any useful technology can be abused. The more "useful" it is... the more "abuseful" it will be in the wrong hands.
HitPaw seems useful... probably more than good enough to "fix" some bad photos.
But also seems dangerous, if it's used to "prove" things that it just does not prove at all. People tend to believe computer-generated answers when they don't really understand the question.

I used to have a rubber-stamp for computer printouts... back in the days when everybody picked up their printouts from the centralized "main-frame" printer... the stamp read:
"This document was produced by a computer. It is therefore not to be disbelieved, or questioned in any way."

I used to be a systems engineer, but back before personal computing was common (the beginning of the IBM desktop and compatibles), I was working on my first master’s degree. I saw all the potential of integrating all the new technologies to come up with bleeding edge (beyond leading edge and top of the spear deployment) capabilities. Some of my courses made me realize that there were also potential risks to moving too fast. Those risks usually dealt with the misuse of technology.

This is a case of bad people finding a use for a useful, good technology. The good thing about that is the fact that if you know how the bad people are able to take advantage of the good things, you can find ways to thwart them. Sensors can be tricked and outcomes can be changed. Understanding the criteria in the code could even enable you to be invisible to the sensors.

Key words!

Tencent is essentially CCP (Chinese Communist Party)

Facebook's facial recognition identified a "friend" as someone else!