Carnegie Mellon-Based Startup Wins DARPA’s $2 Million AI Security Challenge

DARPA’s Cyber Grand Challenge was an all-out melee featuring seven autonomous computer systems for a grand prize of $2 million. The contest, which was the first of its kind, took place in Las Vegas and lasted for a grueling 10 hours and 95 rounds, as the automated security systems competed to identify, exploit, and patch the most security vulnerabilities.

Carnegie Mellon University’s ForAllSecure team emerged victorious after its autonomous system, dubbed “Mayhem”, decisively demonstrated that it could patch software bugs most effectively. “Xandra”, which was built by researchers at the University of Virginia and GrammaTech took home the second place prize of $1 million, UC Santa Barbara’s “Mechafish” took home $750,000, according to The Verge.

While the winning machines were able to identify and mitigate vulnerabilities in fractions of a second, the biggest issue turned out to be simply keeping them running, The Verge reports. Even Mayhem fell silent and ceased working during some rounds of the challenge, earning it a score of zero.

In a statement to PR Newswire, Jim Garrett, Dean of CMU’s College of Engineering said “this is a shining moment for a startup born at Carnegie Mellon: We couldn’t be more proud of ForAllSecure for applying its vision to the development of cutting-edge technology that addresses the global issue of security.”

DARPA instituted the Cyber Grand Challenge in order to encourage and identify state-of-the-art AI systems that could quickly and effectively deal with software bugs on a massive scale. Network attacks are a pressing security matter that grow more dire each day due to the expansion of the Internet of Things. Billions of new connected devices are expected to be made and sold in the coming years, exponentially increasing the occurrence of and opportunities for software bugs.