The Defense Advanced Research Projects Agency, or DARPA, on Wednesday announced a new competitive grant program to design AI-based tools that can find and fix bugs in commonly used software.

The program will be a collaboration between DARPA and four top AI companies: Anthropic PBC, Microsoft Corp., Google LLC, and OpenAI Inc., Arati Prabhakar, director of the White House Office of Science and Technology Policy, told reporters at a briefing Tuesday. The companies will provide their platforms as well as expertise to conduct the competition, Prabhakar said.

The contest is being launched at the annual DEFCON hacking conference starting Wednesday in Las Vegas. The program will award about $20 million in prize money, including a top prize of $4 million, said Perri Adams, the program manager at DARPA who will oversee the contest. DARPA also will fund seven small businesses each with $1 million to help them compete in the first phase of the competition, Adams said.

The winning team must “build a system that can rapidly defend critical infrastructure code from attack,” Adams said, referring to software code used by companies in critical sectors including chemicals, water, transportation and others. “We want to create systems to automatically defend any kind of software from attack, from those used in commercial industry to those running on life saving medical devices.”

The semi-final stage of the contest will be held at the 2024 DEFCON event and the finalists will be chosen at the 2025 event, Adams said.

DARPA has used such challenges and contests to advance several critical technologies including self-driving vehicles and underwater drones. Technologies that emerged from those competitions have gone on to widespread commercial uses as well as design of military systems.

DARPA launched a cybersecurity challenge in 2014 to create automatic defensive systems capable of reasoning about flaws, formulating patches and deploying them on a network in real time. 

“The challenge is critical in bringing together the cutting edge in automatic software security and AI, which will empower our cyber defenses by being able to quickly exploit and fix software vulnerabilities,” Anne Neuberger, the White House deputy national security adviser responsible for cybersecurity, told reporters at the briefing.

The latest competition is intended to advance cybersecurity efforts by using artificial intelligence systems that can identify security vulnerabilities in software and fix them before criminals and spies belonging to adversarial countries identify and exploit those weaknesses, an administration official said during the briefing, speaking on the condition of anonymity.

While criminals and spies also are aiming to use AI tools to “more quickly find vulnerabilities to accelerate development of malware,” the goal of the DARPA contest is to “catalyze a larger community of defenders to use AI for defense,” the official said.