No

I will make the assumption that the universal translator generates a consistent mapping from words, expressions, or idiom in one language to another. I would also assume that it uses samples of language to map potential meanings onto any phrase or sentence, and through some super-advanced algorithm, quickly discerns the meaning of what is spoken.

The problem with cryptography, in this context, is that there is no direct mapping of a word (or expression, or idiom) from one language to another. If I say something, and public key encrypt it, and then you say the same exact thing and encrypt it, those two statements might not be the same thing! Their encoding depends on the user's key. Furthermore, if I say something today, and then the same thing tomorrow, when encrypted those won't necessarily be the same thing either!

The problem with encryption, is that there is no consistent 'language' to be translated. There cannot be a map of a concept like 'blue' to a discrete subset of an encrypted message. 'Blue' could potentially be encrypted any way, depending on what specific cypher is being used that moment by that user.

Therefore, based on the assumptions of how a Universal Translator works that I started with, the Universal Translator will not be able to translate an encrypted message.

An encrypted data stream is statistically indistinguishable from a purely random stream, at least a data stream encrypted by a good encryption algorithm. Since the encrypted stream cannot be distinguished from a random stream, there is no structure on which a universal translator could work. If the encrypted data stream can be distinguished from a random stream then the encryption algorithm is broken; we should expect that when we will be smart enough to build a universal translator we will also be bright enough to write decent encryption algorithms.

One important aspect that all naturally occurring (and some invented) languages have in common is that it has native speakers who learned the language by listening to it. I won't go so far as to say that the languages were "designed" to be learnable, but any language that isn't human-learnable wouldn't last very long.

Encryption, on the other hand, is specifically designed so that a listener who doesn't know the encryption will be unable to decipher it.

A translator that is designed to be able to learn like a human (no matter how much accelerated) won't be able to crack encryption (and the encryption being cracked is crackable by one or more of its algorithms) unless the ability to crack encryption is specifically included, just by virtue of this difference in "design" of language and encryption.

EDIT: This doesn't even go into the technical side of why encryption is difficult to crack, which is a much bigger question (and mostly irrelevant once you consider the above).

The answer is unfortunately simple:

There are no universal translators in real life, so the capabilities of a fictitious universal translator in a story are 100% defined by the author's choice of such capabilities.

Thus the answer to this is plain and simply "whatever you want it to be." That being said, there would be some requirements on a UT that can decrypt encrypted text.

• It needs to be effective at reading between the lines to find the true meaning. What you are describing would only make sense if the translator tries to find the real meaning of things, not just what is said. A translator that can unmangle enigma encoded messages would be able to read between the lines for a hapless male character who is obliviously missing the hints his wife is giving him. (come to think of it... can you make me one of these!?)
• The mere fact that it is possible to translate such meanings should suggest that there is some degree of commonality between all languages. In linguistics, it is not known whether such a commonality exists. Chomsky famously suggested that the concept of recursion was universal to all human languages -- a so called "universal grammar." Strange human languages such as the Pirahã language make it difficult to argue this commonality exists in humans. Commonality between sapient species is beyond that, and in the realm of the author.

No.

This extends on some other answers going into more details, but the short summary is that every language by necessity has patterns. A grammar, a fixed vocabulary, semantics, all of that. A well-encrypted data stream should not have any meaningful patterns, it is indistinguishable from noise or randomness.

A universal translator by definition does not have a database of languages (it wouldn't be universal, or the database would have to be infinite), but rather "learns" a new language it encounters. The only way to do that is to somehow (handwaving) decipher the patterns in the new language. On an encrypted data stream, it would not have any patterns to identify, thus it cannot translate/decrypt it.

Even shorter answer:

No, translation and decryption are completely different processes with different rules and methods, and a system capable of one isn't automatically capable of the other.

Finally, answer with caveat: When you go into the realm of very, very simple encryption (Ceasar ciphers and other substitions), then yes your translator would be able to translate encrypted messages, because these primitive ciphers don't hide the patterns. Which is exactly why they are so trivially easy to break that you can do it with a pen and a sheet of paper.

in order to translate nuance, subtext and handle contextual meaning a universal translator would have to be an AI. capable of actually understanding the language not just an algorithm. Such an AI would be able to break the simplest encryption (substitution, ciphers, ect) relatively easily, but more advanced encryption would just as hard for it as it would for humans.

If you want really hard science ideally both sides would need their own UT which would communicate together at high speed to learn each language. simply not having one on each side would make it much harder and require cooperation.

This sounds like a sword and armour problem.
A sword that can cut through any armour forces the design of an armour that can't be cut through by any existing sword, which forces the invention of a sword that CAN cut through that armour, which forces the design of better armour,....and so on.

Similarly, a translator that can decrypt encrypted information forces the design of better encryption algorithms. Now, unless the translator is specifically designed as a snooping tool, that's where it stops, otherwise you get an encryption arms race.

The answer, for a legitimate device, would be NO.

One thing that wasn't mentioned yet is that a universal translator would be perfect for codewords and code phrases.

A codeword or phrase is basically a substitution for a normal word or phrase, for example a "tank" could be called a "can" and killing someone could be "taking out the garbage".

A universal translator could pick that up as a local dialect, synonyms or sayings and then automatically start translating them not even knowing it was meant as a code.

IT-guy here

Depends on the encryption-scheme in question. For keyless encryption-algorithms it's pretty trivial, since every word can be directly translated into an encrypted word and vice versa. As an example one could name ROT13.

For encryption with keys, things become at least hard to impossible. Consider for example the Caesar-Cipher, which is basically ROT13 with an arbitrary rotation-offset. The issue here is obviously that we need the key to get a proper result. We can of course guess the key, but this requires a way of distinguishing the correct output from rubbish that gets produced when we translate with a wrong key. This way may exist for certain inputs, but there is as well quite a good chance, that another key results in another text that is valid as well.

For modern encryption-algorithms this becomes even worse. Caesar-Cipher has a key between 0 and 26. AES256 encrypts with a 256bit long key. So on average we need 2 ^ 256 / 2 = 2 ^ 255 guesses before we get the correct key. With a quite good chance of finding a few other keys that output other "correct" output as well (correct in the sense that the UT understands it). In addition to that, the computation will take for ever with current super-computers and the worldwide per year produced energy would only be sufficient to test out a fraction of the keys. And at that point we haven't even done any testing whether there is a valid output.

So in short:
For keyless encryption-schemes: sure, it's just a simple mapping. For encryption using a key: theoretically yes, but the encryption needs to be extremely weak and you need to be lucky in terms of the encrypted text to get a definite output. For any stronger encryption you need an enormously strong UT (far beyond what is possible with state-off-the-art resources). Additionaly technology moves also on for encryption. E.g. there are already public-key-encryption algorithms in the making that are resistant to quantum-computers. You'll never be capable to build a UT that is able to translate ciphertext that was translated with state-of-the-art encryption, simply because that's exactly what encryption is supposed to do.