Once up a time, in 1882 and again in 1917, a theoretically perfect cipher was developed. That it was theoretically perfect was proven in a classified report released by Claude Shannon in 1945 which became publicly available in 1945. The system in question is the classic One-Time Pad.
While "perfect" it is extremely problematic in practice. Compared to modern ciphers which can be implemented with comparatively short keys, a one-time pad system requires a key as long as the text to be encrypted. The contents of the one time pad additionally must be both truly random and unknown, so you can't just consult your copy of the classic A Million Random Digits with 100,000 Normal Deviates.1 Additionally once a portion of the pad has been used to encrypt something, it must never be used again. Further presenting obstacles to the secure use of one time pads is the need for a physical exchange of the pads between the communicating parties, an electronic exchange already drops the security of the pad to merely being as secure as whatever encryption protects the pad in transit, which defeats the point of having a one time pad. Marcus Ranum has a decent FAQ on the subject of one-time pad usage.
Practical implementations of one-time pad systems have been broken numerous times over the course of their history. During World War II a number of German messages were broken when the random number source they used was found to actually be producing predictable output. During the Cold War reuse of some older one-time pads by the KGB created circumstances where British and American Intelligence agencies were able decode a number of messages old and new. There have likely been an untold other number of instances where pads had been surreptitiously copied, stolen, or replaced.
You do have a copy of this don't you? ↩