At Amazon, responsible AI development includes collaboration with leading universities to promote breakthrough research. When we acknowledge that many academic institutions lack resources for major studies, we transform the landscape with the Amazon Nova AI challenge. While Amazon Nova Ai Challenge will explore different facets of generative AI (Gen AI), this year’s challenge is centered on “trusted AI: Advancing Secure, AI-Assisted Software Development to build more secure, more reliable applications.”
“This challenge is really unique,” says Dr. Ruioxi Jia, an assistant professor and faculty advisor for the Virginia Tech Team. “Advanced AI research typically requires access to massively trained models or open models with open weight. But unbuved, open weight model often does not have the right level of performance. Amazon is dramatic to lower the research barrier by giving Akademia hitherto real world scenarios.
AI revolutionizes software development by automating boring, yet important tasks, such as updating software updating that frees up teams to focus on innovation. For example, by integrating Amazon Q’s code transformation into Amazon’s internal systems, the team was able to reduce the time it takes to upgrade a Java application to Java 17 from what is typically 50 developer days to just a few hours. This saved an estimated equivalent of 4,500 developer years of work and generated estimated $ 260 million. In annual efficiency gains. As AI becomes more integrated into coding processes, it will inevitably bring new security challenges. By proactively tackling these risks, trust and security are prioritized from the start.
In a “Tournament Format” format “-Teams Teams Teams Teams Teams (DEFENSE) Teams and Five Red (ATTACK) -teams-Teams-Talt in four sequential tournaments to strengthen AI-based secure software development. Each defense team’s code generary model faces all five red teams that will investigate vulnerabilities and deficiencies using automated techniques. -The team built a custom model on AWS Trainium -hardware for the challenge to enable open, collaborative research in safe software development.
The first tournament started in January 2025, when a final round was live in June 2025. All teams will publish research articles that describe their methods and findings, ultimately to improve the user experience, prevent abuse and enable more secure use of AI for software development. The progress of the challenge will contribute to the wider field with responsible AI development in code generation and beyond.
“Research is often very lonely,” says Atharva Naik, the student’s head of Carnegie Mellon University team. “Here we actually compete with other research teams and try to follow their progress and without making them in real time. The tournament format also runs teams to implement and test their strategies quickly without being spoiled in a solution, Naik emphasizes.
The selected teams in each category:
Model developer team
- Carnegie Mellon University
- Columbia University
- Czech Technical University, Prague, Czech Republic
- University of Illinois Urbana-Champaign (UIUC)
- Virginia Tech
Red teams
- Nova School of Science and Technology, Lisbon, Portugal
- Purdue University
- University of California, Davis
- University of Texas in Dallas
- University of Wisconsin Madison
Organizers and participants agree that tournament format has proven to be very motivating. The attack and defense systems must be effective against five opponents, and each team gets better asy face against different opponents in each tournament.
“Together we have the best and brighty from Academia for not only competent, but collectively tackle one of the most important problems in the real world of generative AI-Safe and Safe Software Development,” said Rohit Prasad, SVP for Amazon Artificial Intelligence. “We have designed this challenge as a unique, fast pace tournament to accelerate academic research for practical use. In the Amazon tradition, I look forward to the competing teams that work hard, have fun and make some story along the way to the final.
“This challenge exemplifies our commitment to promoting responsible AI development and security,” said Steve Schmidt, head of security officer, Amazon. “By working with universities, we knock on a well of fresh ideas and cultivate future AI security leaders. This initiative goes beyond theoretical research – it is about developing new ways of identifying safety vulnerability and protecting against threats that can be used directly for generative AI coding assistants. I can’t wait to see what students invent and share their research.
Each team receives $ 250,000 in sponsorship, monthly AWS credits and the chance to compete for top prizes. The winning Red Team and Model Developer Team each receives $ 250,000 (must be divided among students), where team in second place earns $ 100,000. Included scholarships, $ 700,000 in prizes and AWS credits, the total investments in Teams EXCTED $ 5 million.
To gather the best
Amazon organizers are reviewing over 90 suggestions to choose the last ten teams to compete in the challenge. According to Michael Johnston, an applied science leader at Amazon, who oversees science and the technique behind the challenge, it was a tough decision and the harness had to bring a host of unique and practical ideas to the table. Becuse Each team will compete against more opponents, they need to be ready with more strategies. And because their opponents are constantly adapting, these ideas must show creativity and adaptability.
The challenge is by nature interdisciplinary at the intersection of responsible AI, Gen AI, Security, Conversation AI and Automated Software Development. As such, it has bushes together teams with expertise across several fields of study that bring different talents and prospects to the competition.
From theory to practical solutions
Amazon Nova AI Challenge encourages teams to approach problems through a more pragmatic lens than what is often used in academic research. Virginia Tech’s Jia points out that academic literature tends to focus on theoretical problems with a bias against complicated solutions. That’s not what is needed here. The challenge helps to frame the problem in a way that is beneficial to real people, says Jia. In her conversations with Amazon scientists, she says it is clear that they are impressed with solutions that are too complex. “They tell me? “It can be eye -opening and transforms my research philosophy a bit.”
The teams also have access to a resource level that is not often available in academia. “We are trying new things that would otherwise be very hard on an academic budget,” says Naik. His team has generally worked with smaller data sets and has not had an opportunity to train LLMs or run more experience. Kobza was similarly impressed during bootcamp. “We go instant access to AWS Trainium chips, a family of AI -chips -purpose building of AWS for AI training and inference to deliver high performance and at the same time reduce costs. It’s really amazing, especially for a team from a small university in the Czech Republic where we do not have the opportunity to access this powerful hardware.
“The challenge is in line with our research interests really well,” says Professor Xiangyu Zhang of Purdue, an adviser to his school’s team. “And we make the extra step in the competition. We attack from all corners.
Keep an eye on updates on the team’s progress and coverage of the final in June 2025.