TLDR: Team Atlanta, a consortium including Samsung Research, Georgia Tech, KAIST, and POSTECH, has won the prestigious DARPA AI Cyber Challenge (AIxCC) at DEF CON 33, securing a $4 million grand prize. Their AI-powered technology autonomously detects and fixes software vulnerabilities, marking a significant leap in cybersecurity, particularly for critical infrastructure and open-source code.
LAS VEGAS – In a landmark achievement for artificial intelligence in cybersecurity, Team Atlanta, a collaborative effort featuring experts from Samsung Research, Georgia Institute of Technology, the Korea Advanced Institute of Science & Technology (KAIST), and Pohang University of Science and Technology (POSTECH), has been crowned the winner of the Defense Advanced Research Projects Agency’s (DARPA) AI Cyber Challenge (AIxCC) at DEF CON 33. The team was awarded a $4 million grand prize for their groundbreaking AI-powered system capable of autonomously identifying and patching software security vulnerabilities.
The two-year competition, which culminated on Friday, August 8, 2025, aimed to accelerate the development of AI tools that can automatically secure the critical code underpinning prominent systems globally, including those in vital infrastructure sectors like hospitals, power plants, and water systems. Andrew Carney, AIxCC Program Manager, emphasized the profound impact of the challenge, stating, “The world is different because the AI Cyber Challenge (AIxCC) has fundamentally changed our understanding of what is possible in terms of automatically finding, but more importantly, fixing vulnerabilities in software.”
Team Atlanta’s winning technology, which notably integrated traditional threat hunting tools with advanced AI, demonstrated superior performance throughout the rigorous final round. Competing against dozens of teams, they excelled in finding and proving vulnerabilities, generating patches, and pairing vulnerabilities with patches, achieving the highest rate of accurate and quality submissions. DARPA officials noted that Team Atlanta significantly outperformed its competitors, earning top marks in nearly all categories.
During the final competition, teams were tasked with identifying and generating patches for synthetic vulnerabilities embedded within a massive 54 million lines of code. The finalists collectively discovered 54 out of 70 synthetic vulnerabilities, achieving a 77% success rate, and successfully patched 43 (61%). Remarkably, the competition also led to the discovery of 18 previously unknown, real-world software vulnerabilities, which are now being disclosed to developers. On average, the AI models developed by the teams were able to patch flaws in just 45 minutes.
Stephen Winchell, Director of DARPA, highlighted the urgency and scale of the problem AI is now addressing. “We’re living in a world right now that has ancient digital scaffolding that’s holding everything up… And the reality is [that] it is a problem that is beyond human scale, and it’s a critical problem that we need to solve right now.” Carney added, “They’ve shown that they can patch real software quickly, scalably [and] in a cost-effective way, and those tools are yours to use. There is no excuse not to leverage this flavor of automation. And it will only get better. This is the new floor.”
Taesoo Kim, a professor at Georgia Tech and leader of Team Atlanta, expressed the team’s commitment to advancing research, announcing their decision to donate a significant portion of their winnings back to Georgia Tech. Kim also noted the potential for their ‘AIInspect’ security technology, dubbed ‘Next-Generation AI Cybersecurity for Smartphones,’ to expedite the rollout of self-defending Samsung products, affirming Samsung’s position as a top AI security technology provider.
The competition also recognized other leading innovators: Trail of Bits, a New York City-based cybersecurity firm, secured second place with a $3 million prize, and Theori, a team of AI researchers and security professionals from the U.S. and South Korea, took third place, receiving $1.5 million.
Also Read:
- Team Atlanta Secures $4 Million DARPA Prize in Groundbreaking AI Code Security Challenge at DEF CON
- Researchers Unveil Zero-Click Prompt Injection Vulnerabilities in AI Agents at Black Hat Conference
A key outcome of the AIxCC is the commitment to open-source distribution. The cyber reasoning systems (CRSs) developed by the finalists will be released as open-source software, with four models already available and three more expected soon. This initiative aims to accelerate the adoption of these advanced AI models to enhance cybersecurity across critical infrastructure and beyond.
