Adam Tauman Kalai is an American computer scientist who specializes in artificial intelligence and computer science. He was a researcher at OpenAI from 2023 to 2026 and is a cofounder of the Institute for Responsible Superintelligence (RESI). His affiliations have included the Institute for Responsible Superintelligence, OpenAI, Microsoft Research, Georgia Tech, and the Toyota Technological Institute at Chicago. Kalai attended Harvard University and Carnegie Mellon University, authoring the 2001 work "Probabilistic and on-line methods in machine learning" under advisors Avrim Blum and Santosh Vempala. He is connected to Ehud Kalai, and his personal website is https://kal.ai/.

Kalai graduated from Harvard University in 1996 with a BA in computer science, and went on to receive an MA and PhD, both in computer science, from Carnegie Mellon University in 1999 and 2001, respectively, under the doctoral advisement of Avrim Blum.^[1] After graduation, he completed his postdoctoral research at the Massachusetts Institute of Technology under Santosh Vempala until 2003.^[2] Kalai was a faculty member at the Toyota Technological Institute at Chicago from 2003 to 2006, followed by a stint as an assistant professor at the Georgia Institute of Technology from 2007 to 2008.^[3] He joined Microsoft Research in 2008, subsequently moved to OpenAI in 2023, and left OpenAI in 2026 to cofound the Institute for Responsible Superintelligence (RESI).

Kalai is known for his algorithm for generating random factored numbers (see Bach's algorithm), for co-inventing the cooperative-competitive value (coco value), for efficiently learning mixtures of Gaussians, for the Blum-Kalai-Wasserman algorithm for learning parity with noise, and for the intractability of the folk theorem in game theory. In 2016, Kalai was a co-author on a widely cited paper on identifying and reducing gender bias in word embeddings, which are representations of words commonly used in AI systems. In 2026, he coauthored a Nature paper on hallucinations in large language models.