The potentials at the NIST site (www.ctcms.nist.gov/potentials) have references to the papers containing the fitting information. Almost all of the files on the site came from, or were approved by, the developers themselves. Thus they probably accurately reflect what is in the references. Read the reference(s) and other papers that have used that potential to make sure it works well for the problem you care about. There may be many potentials available for a particular system, and you have to use your judgement in selecting one.
For example, if you care about solid-liquid interface simulations and a potential does not reproduce liquid properties well, then you may not want to use that one. If a potential was not fit to high-temperature properties, it may not behave well at elevated temperature. If the form (equation) of the potential does not capture physics that is important for your problem, it may not be the best choice, even if it works well in another application.
With any of these potential files, you should know where the files came from (who generated or modified them) and assure yourself that the potentials as represented in the files give you the expected answers. This means that you should test to make sure that you reproduce some or all of the properties reported in the papers before you use the files for real research.
Also, if the file lists parameters, make sure the parameters in the files are the same as the ones in the associated papers. If they aren't, you should find out why. Did the paper have a typo? Does the potential file have a typo? Is the listed reference actually the correct one? Did somebody modify the file without documenting it? Is something else going on?
Be aware that sometimes file/potential formats change, and LAMMPS has flags for some potentials that need to be set properly to reproduce the published results. One example is the augt1 flag for MEAM. You should read through the options on the LAMMPS potentials pages to check that the default flags are the proper ones. Then test to make sure you reproduce properties from the paper (e.g., cohesive energy). An incorrect flag can lead to subtly wrong answers that undermine your work.
Since your research results depend on the interatomic potential you use, you need to be able to defend your choices beyond "well, this is the file that came with LAMMPS."
Cheers,
Chandler