Several ways to tackle this question. And it’s a very good question.
Both consonants and vowels in the IPA are defined, not against a word of a language (they can indeed change), but against an articulatory gesture. Because people’s oral cavities are pretty much the same, that works. [ç] is defined as a Voiceless palatal fricative, not as the <ch> in German ich (Which dialect of German?), or as the χ in Greek όχι (Which dialect of Greek? Which period, for that matter?)
For vowels, it’s only slightly more complicated. There is indeed a continuum of articulation, of where you place your tongue in your oral cavity, to produce vowels. But:
- There are eight reference vowels, the Cardinal vowels, which are defined through articulatory gestures as signposts for the rest of the IPA vowels;
- There is an articulatory space defined for vowels —
- —but the two dimensions of the space can actually be plotted precisely based on the formants of the vowel—the first and second peak of the vowel sound in a spectrogram. See Vowel diagram. So you can run a spectrogram on vowels, and get a precise plot of vowels like this:
—the classic vowel trapezium plots F1 against F2–F1.
In any case, the IPA is not intended for a precise plot on the trapezium. It’s intended for a close enough area of the mouth, that sounds the same to listeners as other speakers’ vowels, and that sounds different to other vowels of the same language. Two speakers’ vowels are not going to plot to exactly the same place. The IPA is used for linguistic transcription, which is based on contrast; it is not intended to substitute a spectrogram.
So if people start pronouncing cat slightly differently, but it’s still in the general area of [æ], few linguists are going to care. Particularly if the [æ] is still nowhere near the [ɑ] or [ɐ] or [a] of that dialect of English. If it does, why, we’ll relabel it to the other area of the vowel chart that it’s moved to. /kʰæt/ used to be /kʰat/ (and still is in some dialects of English).
If you’re one of the few linguists that do care, you’re not looking at the IPA anyway. You’re looking directly at the spectrograms.
You’ll see that the IPA vowel chart allows 7 degrees of height, and 3 degrees of backness; diacritics allow both to be tripled. The 7 degrees of height is new-fangled, and I’m not sure whether symbols like Close-mid central unrounded vowel [ɘ] have really been used seriously. Even with the diacritics ̟ ̱ ̝ ̞ , I doubt most transcriptions have ever bothered with more than four degrees of height.
So the IPA doesn’t need to make more symbols to keep up: it’s got an oversupply if anything. (The Labiodental flap symbol /ⱱ/ is the first new IPA consonant in decades.) You just make sure the definitions of the symbols are independent of words in any one language.