The regulator should be short-circuit protected, so not burn out very easily. Have you bought the regulators from a reliable source, there are lots of counterfeit LM323's around and maybe also LM317K?

Powering up the MPU on the bench will enable you to measure current consumption, should be about 0.3 - 0.5 amps from the 5V line.

Check that your CPU board or connector does not have a short between +5 and +12V, the regulator is usually quite robust but does not tolerate external higher voltages at output.

Maybe removing the case top with a hacksaw from the original part and one of the suspect counterfeits could reveal difference in die size.

Just checked my latest batch of spare LM323K's from eBay. All have date code 0452, and all the markings could be wiped off with some contact spray. Bummer!

And yes, we have seen the 2N2222 or something like that in a TO-3 cased 2N3055. And also the 10000 uF electrolytics, with a Chinese brand 1000 uF inside the can.

I guess I will make a test bench for my LM323s, run them for an hour properly heat sinked at 2A and 10V input, making 10W power dissipation - about the worst case in WPC driver boards.

Fortunately I have not installed those eBay regulators in any of my friends boards yet.

lm323 (resized).jpg

Yes, the lid is off centered in random direction in nearly all of the 10 pcs I have.

And what's even worse, the parts are either faulty, or are not LM323's. I will test them again as LM317's, because they look exactly like OP's fake LM317K. Maybe I will open one and try to look at the die with microscope, if there are any markings.

Does not look like a 3 amp die or bonding wires to me.
lm323-3 (resized).jpg

I might have one failed original part in my gameroom trash bin. Will open that when I get there.

So here. A Linear Technology dated 1992. The die size is not so much larger than the fake one, but there a two pairs of heavier bonding wires, going to different places on die, probably to minimize hotspots. Or there might be two output transistors in parallel.
lm323_92 (resized).jpglm323_92_1 (resized).jpg

Sure!

The cut open parts and cases I posted are both LM323K's, first the fake then the original. The first picture by OP is his LM317 case.

My fake parts do not work as LM323, but not as LM317 either. I don't know what the die actually is.

Now that's a good idea!
20220825_130257 (resized).jpg

Clearly, not a transistor either.

A working original LM323 connected to tester in same way resulted in "No, unknown or damaged part".

The component tester identified some of my 7805's as Darlington transistors with high Vf of about 3V. Some 7805 were identified exactly like the fake LM323, with normal diode between output and ground, and a high drop diode lookalike.

So it is very possible that the fake LM323's were made using 7805 die, but something went wrong, because they take almost 60 mA input current with output unconnected, and voltage at output is under a volt. I had 0.1 uF caps to ground at input and output so it was not oscillating. Just for fun, I might try testing all 10 pcs, but into trash they go even if some would imitate a 7805!

Edit: All tested. Two out of 10 were completely open (missing/broken bonding wires or forgot to install die?). The rest 8 were all alike. With 8V input, quiescent current was 60 mA and output 0,45V.

I tried swapping the input and output, just in case the faker had cross-connected bonding wires. Then I tried the negative voltage regulator pinout and connection. No luck.

Be careful what you buy, especially if it is an obsoleted component. With the LM323/LM317 regulators, especially watch out for the NS logo and 0452 manufacturing date.

"Sometimes you get more than you bargained for. Beware of unusual parts... in the Twilight Zone!"