It looks like the plural of 'prefix' is 'prefixes' - while I would expect it to be
'prefix' => 'prefices' like
'matrix' => 'matrices' or
'index' => 'indices'.
Is 'prefix' an exception to the rule? If so, why?
The plural of the Latin word matrix is matrices, and the plural of the Latin word index is indices. We took the singular forms of these words from Latin unchanged, and the same goes for the plural forms (unchanged in the spelling, anyway; the pronunciation has been anglicized).
But prefix is not technically a Latin word. It is derived from the Latin word praefixum, and the plural of that is not praefices, it's praefixa. Since there is no Latin word prefix or Latin plural form prefices, the English plural form is just prefixes, made using the singular prefix + the normal English plural suffix -(e)s.
From the Online Etymology Dictionary entry on prefix (n.):
1640s, from Latin praefixum, noun use of neuter past participle of praefigere "fix in front, fasten on before," from prae "before" (see pre-) + root of figere "to fasten, fix" (see fix (v.)).
Compare this to the entries on the origins of
late 14c., "uterus, womb," from Old French matrice "womb, uterus," from Latin matrix (genitive matricis) "pregnant animal," in Late Latin "womb," also "source, origin," from mater (genitive matris) "mother" (see mother (n.1))[...]
late 14c., "the forefinger," from Latin index (genitive indicis) "forefinger, pointer, sign, list," literally "anything which points out," from indicare "point out" (see indication). Meaning "list of a book's contents" is first attested 1570s, from Latin phrases such as Index Nominum "Index of Names," index expurgatorius "specification of passages to be deleted from works otherwise permitted."[...]
Sven Yargs left a comment mentioning another English word derived from Latin that has a singular form ending in "x" and a regular plural form ending in "xes": reflex, which like prefix derives from a Latin past participle of the second declension rather than from a Latin noun of the third declension, and which has also dropped the Latin singular ending (in this case, the masculine suffix -us).
Some other nouns that do not take a plural in -ices even though the last two letters of the English singular are -ex are circumflex, complex, duplex, multiplex and annex (and of course the monosyllable sex).
Aside from the other -fix words (affix, suffix, postfix, infix, antefix, superfix, subfix, crucifix etc.), I can't find any more Latin-derived nouns ending in -ix that lack (standardly recognized) plural forms in -ices. Such words would be expected to come from sources like Latin nouns ending in -ixus, ixum or perhaps -ixa.
As far as I know, there are no English words ending in -ax, -ox, or -ux that commonly take plural forms in -ces, even though there are some Latin words ending in these letters that pluralized this way (e.g. fornax, fornaces "furnace", vox, voces "voice", crux, cruces "cross", dux, duces "leader").
The word syntax and flux don't come from Latin words ending in -x (they are from syntaxis and fluxus), and so the absence of *"syntaces" and *"fluces" as plural forms is expected.
The Latin form of the word climax does end in -x, and so there's no way in theory to rule out the Latin/Greek-based plural form climaces as a possibility for the English plural, but in practice climaxes seems to be the only possibility for the vast majority of present-day speakers. Climaces apparently was used in Kinsey, Pomeroy, and Martin's Sexual Behavior in the Human Male (first published 1948, copyright renewed 1975; accessed through Google Books).
So, I stated a general principle, but of course not everything is as simple as that. Not everyone is aware of this principle, or follows it all the time, and there are complicating factors like analogy. Interestingly, Wiktionary (and apparently the OED as well) does list "prefices" as a ("nonstandard") plural form of "prefix". So it apparently has seen some use.
There are some other (rather odd) possible exceptions in English to the general rule that Latin plural forms are only used when the corresponding Latin singular form is used.
The word antipode, as far as I know, is only ever pronounced /ˈæntɨˌpoʊd/ "ANT-ih-poad"). But the singular form antipode is less common than the plural form antipodes, which coincidentally has an ambiguous written form: it corresponds equally well to the regular English -(e)s plural and the Latin form.
The OED says that the original English pronunciation of the plural form antipodes was probably regular (/ˈæntɨˌpoʊdz/ "ANT-ih-poads"), and compares it to the French word antipode/antipodes (which is formed in the regular way for a French plural). But it is most common nowadays to pronounce this word as if it were taken directly from the Latin plural form, giving /ænˈtɪpədiːz/ "an-TIP-uh-deez".
There seems to have been no (or at any rate, no attested) Latin singular form because the word was always used as a plural (it comes from the same "foot" root as octopus, which has the opposite issue of an unclear Latin plural form). So those who say Latin-based /ænˈtɪpədiːz/ for the plural have to fall back on non-Latin /ˈæntɨˌpoʊd/ for the singular.
The word mores, as in "social mores," comes from the Latin mores "customs" (a plural word) and is prescriptively pronounced with two syllables, not one (either /ˈmoəriːz/ "MORE-eez" or /ˈmoəreɪz/ "MORE-ayz" is generally considered acceptable; I prefer the former but dictionaries suggest the latter is more common).
The dictionaries I have looked at say this word only possesses a plural form. The Latin singular form of mores, mos, is never used in English, but you can sometimes see people use more as a singular in phrases like "a social more". I assume this would be pronounced /moər/ "moar" (though since I have only encountered it in writing, I don't know for sure).
Although I would not recommend the use of singular more (since to me it seems to indicate ignorance of the etymological history of the plural form mores, and it has not yet become legitimized by universal usage), it would be another example of a word like antipodes/antipode where the plural is commonly pronounced as a Latinate form, but the singular cannot be.
Latin had a word magus, with the regular plural magi. These forms have been taken into English as magus (pronounced in English /meɪgəs/ "MAY-guss") and magi (pronounced in English /meɪdʒaɪ/ "MAY-jye"), but these words are somewhat obscure. The anglicized form mage (pronounced /meɪdʒ/) has become fairly popular in fantasy fiction. And apparently some fantasy writers pair the anglicized singular form mage with the Latin plural form magi, although I believe it's more common for mage to get a standard English plural mages /meɪdʒɨz/.
This seems to be a case where some people use the Latin plural just because it is fancier, and therefore more fun(?) to use. Also, the Latin plural form magi is more well-known than the Latin singular form magus because many people are familar with the term "Magi" in reference to the Christian Biblical Magi.
There is also a Japanese manga series called "Magi: The Labyrinth of Magic", and in English translations of this work, it seems the term "magi/Magi" is used as a singular noun with an invariant plural form. This feature may be transferred from the original Japanese, since Japanese does not inflect nouns for grammatical number.
Some words ending in a non—primary-stressed syllable followed by /s/, such as process, bias, auspice, interstice, have plural forms ending in -es that are pronounced by some speakers with /siːz/, as if they were Latin/Greek plural forms. But none of the singular forms of these words comes directly from Latin, and although auspices exists in Latin as the plural form corresponding to the singular auspex, this is just a coincidence: the English noun auspice comes from the distinct (but related) Latin noun auspicium, which doesn't mean the same thing as auspex.
I give a somewhat longer list of words where this pronunciation may be possible in my answer to the question What is the correct pronunciation of the word “processes”?