DARPA eyes transition of AI Cyber Challenge tech to 'widespread use'

DARPA has concluded its inaugural AI Cyber Challenge competition that used artificial intelligence to rapidly identify and fix software vulnerabilities, with Team Atlanta from Georgia Tech and Samsung Research taking first place.

The competition addressed the growing challenge of finding and patching cyber vulnerabilities using traditional methods, which are slow, expensive, and depend on limited human resources. Meanwhile, adversaries increasingly use AI to enhance their cyberattacks. The winning AI systems demonstrated remarkable efficiency by discovering 54 synthetic vulnerabilities in 54 million lines of code and successfully patching most of them, while also uncovering 18 real vulnerabilities in actual software.

• AI systems completed vulnerability detection and patching tasks for an average cost of $152, compared to traditional bug bounties that range from hundreds to hundreds of thousands of dollars
• Four of seven finalist systems have been released as open source code, with remaining systems planned for release in coming weeks
• The competition was a collaboration between DARPA and ARPA-H, with top teams winning $4 million, $3 million, and $1.5 million respectively
• Technology will be transitioned for use in federal government agencies and critical infrastructure protection, including energy grids and transportation systems