The Defense Advanced Research Projects Agency (DARPA) has announced its sponsorship of a two-year competition aimed at creating a new generation of cybersecurity tools to enhance software security. DARPA, a research and development agency of the US Department of Defense (DoD), is responsible for developing emerging technologies for military use. The competition, called the AI Cyber Challenge (AIxCC), aims to develop AI-driven systems that can address cybersecurity issues and ensure more secure software. Perri Adams, the program manager for AIxCC, made the announcement on the keynote stage at the Black Hat conference.
Adams emphasized the significance of software in modern life and how it also serves as an expanding attack surface for malicious actors. She acknowledged that recent technological advancements offer promising ways to keep defense one step ahead. According to Adams, the responsible use of AI has remarkable potential to secure software code. However, she stressed that the potential of AI alone is not enough and that it requires a “forcing function” to bring together top figures in AI and cybersecurity to demonstrate how AI can be used for good.
The AI Cyber Challenge will be concluded in 2025 at DEF CON. Participants in the competition will be tasked with designing AI systems that can rapidly identify and fix vulnerabilities in critical code. Adams sees the challenge as an opportunity to use technology to make a significant impact and bring about structural change. She expressed hope that the DARPA challenge would foster incredible innovation.
The AIxCC competition will consist of two tracks: the Funded Track and the Open Track. Competitors in the Funded Track will be selected from proposals submitted to a Small Business Innovation Research solicitation. Prizes for the competition include $20 million for the teams with the best systems and up to $1 million for seven small businesses.
Semifinalists will be announced next summer at DEF CON 2024, and winners will be announced the following year at DEF CON 2025. The top five semifinalists will each receive $2 million and a year to further advance their technology. Their task will be to develop a system that can defend critical infrastructure from attacks.
The AI Cyber Challenge is supported by Google, Anthropic, Microsoft, and OpenAI. The Open Source Security Foundation will serve as a challenge advisor. The competition aims to shape innovation around real-world problems and develop systems that can automatically defend any type of software from attack, from commercial industries to life-saving medical devices.
Michael Sellitto, head of geopolitics and security at Anthropic, believes that AI technology is moving quickly and that software developers are already using AI to write significant portions of code. He sees the potential for AI to interpret or suggest alternatives to code and expects the technology to become proficient in finding and fixing vulnerabilities. Sellitto believes that the AI Cyber Challenge will accelerate these efforts, noting the substantial progress made in the past two years.
The goal of the AIxCC challenge is to develop usable systems that can have a significant impact on software security. Adams envisions cutting-edge technology that can secure software at scale by identifying vulnerabilities, proposing fixes, and replacing vulnerable code with secure code. The success of the challenge is expected to serve as an example of how AI can be utilized to tackle key societal challenges.