Daniel Bernstein (DJB) is a renowned systems programmer and a serious academic cryptographer. Many of the SHA-3 candidates are partly derived from his work, and he designed one of the eSTREAM finalist stream ciphers.
Bernstein, Lange, and Schwabe designed NaCl. NaCl is a high-speed crypto library designed to correct all the things that go wrong with "low level" crypto libraries like OpenSSL. This is a big deal, because OpenSSL is the current "lowest common denominator" crypto implementation and thus forms the basis for most crypto facilities in high-level languages like Python and Ruby. OpenSSL is extraordinarily error prone and itself has a spotty track record.
NaCl provides an ultra-simple interface (crypto_box_+) that handles the details of message signing, encryption primitives public key, formatting, &c.
It's been used by a bunch of projects already (also note that DJB software is somewhat inherently credible; he wrote qmail and djbdns, which have the two best security track records in all of serverside software). A bunch of very smart people have contributed to it.
NaCl addresses a bunch of serious systemic security pitfalls:
* It's hardened, fundamentally, against the "side channel" effects we know about today that allow attackers to measure crypto operations to extract keys. For instance: "NaCl systematically avoids all branch conditions".
* It authenticates messages, in constant time, before attempting to decrypt, which prevents attackers from exploring how the target is handling ciphertext (for instance, this prevents padding oracles in asymmetric and block ciphers).
* It handles the details of getting crypto-secure random numbers in a sane way, and also minimizes places in the design that require random numbers (for instance, DSA requires a random nonce in addition to a strong random key; that's something many people have screwed up).
* It makes extremely well-informed choices about ciphers and algorithms (it's designed by active contributors to the literature).
* It's crazy fast (crazy fast is basically Bernstein's current academic focus).
This thing is simple, it's fast, and it's designed to be safe by people who know what that means.
Start building language bindings for this thing!