I never heard about those (pre-)historical languages before hearing about them in the internet. For instance, proto-slavic would be the ancestor of all slavic languages, and proto-indo-european the ancestor of all indo-european languages (including proto-slavic).
However, is there any evidences that such languages actually existed at some point in history, or are them just an unverified theory made up by linguists ?
This question seems confused. For one thing, you never define what you mean by “such languages”. A precise formulation of the concepts is important.
No reasonable linguist would support the idea that, e.g. the resemblance between English “brother” and Latin “frater” (and contemporary French “frère” etc.) is just due to chance. There are too many parallel cases of words with similar meaning in English and Latin that look very similar if you just apply some (relatively) simply formulated sound laws; we call these words "cognates". (If you’re interested in historical linguistics, you need to familiarize yourself with the concepts of sound laws and cognacy if you haven’t already.)
So, we’ve established these resemblances are not due to chance. The simplest remaining explanation is that these similar words originally come from one source. (There are other conceivable explanations for non-chance resemblances, e.g. “contact caused convergence of these words’ forms”, but that’s prima facie less likely and there’s no reason to suppose that is the explanation in general.)
Obviously, how these words came to be in multiple extant languages from one original source is a complicated issue.
Pretty much everyone agrees that languages, in general, seem to change continuously. For example, there is a language called “English” today, that is a continuation of a language that existed 200 years ago that was also called “English” which was slightly different.
Also, it doesn’t seem to be common for languages to just appear out of thin air (the only examples that I know of of this apparently happening are sign languages, such as Nicaraguan Sign Language).
So it seems reasonable to suppose that the spoken languages we see today did not originate de novo any time recently, but developed somehow from past languages, and probably they developed specifically by processes of continuous change.
If English developed from past languages by continuous change, and Latin developed from past languages by continuous change, the simplest way of explaining a body of shared vocabulary is to say that if you go far enough back, the language that was going to develop into English was the same as the language that was going to develop into Latin.
Radically different hypotheses like “mass borrowing of the lexicon from one language to another distinct language” are more complicated and would have to be justified by some kind of evidence of this process occuring. But we don’t have evidence that indicates that English was aways a distinct language from Latin, or that the English words cognate to Latin words were borrowed into English at some point.
Obviously there are disputes about the details, like “tree models” vs. “wave models” but these should not be misunderstood as disputes about the existence of a relationship between languages in the same family; they are just different models of how the languages diverged and diversified over time. Just as how disputes in biology about the mechanisms of evolution don’t affect the consensus that it occured, disputes in linguistics about the mechanisms of language change don’t affect the consensus that Indo-European languages are all related.
Also, linguists are extremely aware that reconstructions, e.g. forms like *bʰréh₂tēr for the common ancestor of "brother" and "frère", don't give a perfect description of an actually existing past language. That is the point bytebuster is making in the comments; nobody maintains that our reconstruction actually existed as a real language at some point in prehistory. The reconstruction, if it is done correctly, represents some information about prehistoric spoken forms that has been preserved in extant languages. It's unfortunate that the "Proto-X" names are often used for both concepts; I believe the hypothesized actual spoken language/dialect group is sometimes distinguished with the name of "Common X" (e.g. "Proto-Indo-European" would be our reconstruction of "Common Indo-European", which is whatever was actually spoken).
I guess another related issue is how variable the common spoken source was e.g. is it more accurate to call Common Indo-European a "language" or a "dialect continuum", but that's just dealing with a matter that is fuzzy even today: the distinction between "language" and "dialect".
Radically different hypotheses like “mass borrowing of the lexicon from one language to another distinct language” are more complicated and would have to be justified by some kind of evidence of this process occuring Sorry, but mass borrowing definitely happens, for instance french->english or chinese->japansese; as well as the creole language. The existance of words with similar roots in two languages doesn't imply one unified langauge ever existed.
Proto-languages are indeed theories, the way that Evolution or Gravity is a theory. Confirmation of theories comes through their fit of the data, as @sumelic argues.
As theoretical constructs, they are unstable, and subject to refinement—as the iterations of https://en.wikipedia.org/wiki/Schleicher%27s_fable illustrate. They are best guesses, and they are driven by trying to account for modern data, rather than as exercises in time travel. That's why the fact that they aren't always particularly pronounceable or typologically plausible is not really the point. They are not meant to be real languages; they are meant to be short-hand accounts of how a bunch of modern languages are related.
An added problem with the historical accuracy of proto-languages is that they only reconstruct what has survived in all the daughter languages; if a feature has died out in all daughter languages, you're unlikely to posit it for their ancestor, because you simply won't know it was ever there (and because you're following Occam's Razor in your modelling exercise.)
Now, because proto-languages are posited to have a historical reality, they could potentially be verified objectively through a time machine—comparing how close a fit the reconstructed proto-language is to what was actually spoken on say the steppes of Ukraine 5000 years ago. Unlike gravity. But like evolution, which would (and does) get confirmed through "missing link" fossils.
And the proto-Indo-European spoken on the steppes of Ukraine 5000 years ago (maybe) might not look all that close to our reconstruction, for reasons @sumelic mentions in passing. The history of IE may be explained by the Wave model more than we'd prefer, with Trubetzkoy proposing the extreme account that there may not even have been a single proto-Indo-European language. Trees are neater models than waves, and that's why we use them; they're not necessarily more historically accurate. But again, the point of proto-reconstruction is to explain regular relations between modern languages; it's not time travel.
So, has there been confirmation? Not a lot, but some. A standard exercise is to reconstruct Vulgar Latin from Modern Romance languages, and to see how close it comes to what little Vulgar Latin we have. Sardinian is pretty critical in any reconstruction, because it preserves stuff from Latin that no other daughter does (which proves how fragile the reconstructions are.) From what I recall, the reconstruction is close, although we have so little Vulgar Latin attested, we rely on reconstruction to make sense of it anyway.
And in fact that circularity gets in the way of the few "confirmations" of proto-languages there have been. You could argue that the deciphering of Linear B confirms the reconstruction of kw in Proto-Greek, for example. But Linear B is such a poor transcription of Proto-Greek, that it would be impossible to decipher without the reconstruction of Proto-Greek in place to begin with. The deciphering of Linear B is so contingent on the reconstruction of Proto-Greek, that it'd be misleading to say it confirms that reconstruction.
Same goes for the one laryngeal that turned up in Hittite: it confirmed that Saussure wasn't just doing silly algebra when he came up with the laryngeals for Hittite, but it really is just one out of three laryngeals, and people embraced the other two laryngeals because they made sense as a reconstruction—whether or not they left any consonantal relics in Hittite.
