Craig Gentry is a member of the Cryptography Research Group in the Security Department in IBM's T. J. Watson Research Center. He obtained his Ph.D. in computer science from Stanford in 2009, advised by Dan Boneh. He also has a law degree from Harvard Law School.
In 2009, Craig constructed the first fully homomorphic encryption (FHE) scheme, which allows data to be processed in arbitrarily complex ways while it remains encrypted, solving a major open problem that had been unsolved for 30 years. FHE allows data processing to be outsourced (e.g., to the cloud) without sacrificing privacy -- in particular, without disclosing the decryption key. With Shai Halevi and others, he has constructed some of the most efficient FHE schemes, including the Brakerski-Gentry-Vaikuntanathan (BGV) scheme, which is available in an open source implementation called HeLib maintained by Shai and Victor Shoup.
In 2013, Craig (with Shai and Sanjam Garg, then a postdoc at IBM) also constructed the first cryptographic multilinear map scheme, a cryptographic tool that is (in some ways) even more powerful than FHE. This led to the construction of the first cryptographic program obfuscation schemes (with Shai, Sanjam, Mariana Raykova, Amit Sahai, and Brent Waters), a major breakthrough that had been thought to be impossible. Unlike FHE, cryptographic multilinear maps and cryptographic program obfuscation are currently too slow to be feasibly implemented and their (in)security is not well-understood; this remains an active area of theoretical research.
Much of Craig's recent work, including FHE and cryptographic multilinear maps, generally falls into the area of "lattice-based cryptography". Unlike commonly-used cryptosystems like RSA and elliptic-curve cryptography, lattice-based cryptosystems cannot feasibly (as far as we know) be broken by quantum computers. (The NSA cited this as a reason to begin transitioning to lattice-based cryptosystems, or to other "post-quantum" cryptosystems.)
Craig has also worked on verifiable computation (VC), which allows a user to outsource a computation (e.g., to the cloud), such that the cloud can provide an extremely concise cryptographic "proof" to the user that it performed the computation correctly. The paper "Pinocchio: Nearly Practical Verifiable Computation" by Bryan Parno, Craig, Jon Howell, and Mariana, showed that VC can be surprisingly fast for a variety of types of computations.
Craig has won several awards for his work, including the Macarthur "Genius" Award (2014), an invitation to present at the International Congress of Mathematicians (2014), the ACM Grace Murray Hopper Award (2010), the ACM Doctoral Dissertation Award (2009), and several best paper awards: IEEE Symposium on Security and Privacy 2013 (for the "Pinocchio" work), Eurocrypt 2013 (for the cryptographic multilinear maps work), Crypto 2010 (for work in lattice-based cryptography), and FOCS 2007 (for an identity-based encryption scheme).