UNIVERSITY PARK, Pa. — Lion-0xA, a team of Penn State students from the College of Information Sciences and Technology (IST) and the College of Engineering’s School of Electrical Engineering and Computer Science (EECS), is one of 10 global finalists in the Amazon Nova AI Challenge, an annual university competition dedicated to accelerating the field of artificial intelligence (AI).
Now in its second year, the challenge was created to recognize and advance students from around the world who are shaping the future of artificial intelligence, according to the event website. The 2026 competition is focused on building agentic AI systems that are trustworthy and functional.
Dongwon Lee, professor and interim head of IST’s Department of Privacy and Security Informatics, leads Lion-0xA as the faculty adviser.
“I’m excited about the opportunity to participate in this year’s challenge and to represent Penn State on a global stage,” Lee said. “Our proposal was selected from hundreds of applications worldwide, which is both an honor and a testament to the strength of our team. We’ve brought together talented students with deep expertise in adversarial machine learning, cybersecurity and natural language processing, as well as prior experience competing in — and winning — major security competitions.”
The competing teams are creating AI agents that can complete multi-step coding tasks similar to what human software engineers do in their daily work. The challenge is to find a balance between improving AI to handle more complex tasks while ensuring it does not create new risks or problems, because as AI agents become more capable, new security challenges emerge.
The participants have access to Amazon’s Nova Forge service for building customized AI models. This infrastructure, not typically available in academic research settings, provides the computational resources needed to train and evaluate large-scale AI models.
In the tournament-style competition, five of the challenge teams will compete as defense teams that develop agent models. The other five teams — including Lion-0xA — will compete as red, or attack, teams and test those models for weaknesses by attempting to identify potential vulnerabilities.
The Nova Forge service provides teams with model checkpoints at different training stages, allowing them to add their own research and security methods into the models’ foundations. The results are customized models that combine Nova’s capabilities with the team’s specific approach to secure software development.
“What makes this opportunity especially meaningful is the chance for students to work with real-world Amazon datasets and production-level AI systems,” Lee said. “Engaging with industry-scale models and attack-defense scenarios will provide an unparalleled, hands-on learning experience. It’s not just a competition; it’s a rigorous training ground for the next generation of AI and cybersecurity researchers and practitioners.”
The teams gathered Feb. 2-4 in Seattle for bootcamp. Six more tournament rounds will be held from March through August, and the final tournament will take place in September. Each participating team will publish a research paper on their methods and findings from the challenge. Lion-0xA’s original submission, with Lee as the principal investigator, is titled “Autonomous Co-Design and Multi-Agent Red Teaming for Advancing Trusted Agentic AI.”
The teams will present their research, and the competition winners will be announced, at the Amazon Nova AI Challenge Summit in October.
"Being among the top 10 teams globally is meaningful recognition, but more importantly, it is an opportunity to turn serious research ideas into something practical,” said Mohammed Toufikuzzaman, team co-leader and a second-year graduate student pursuing a doctoral degree in computer science and engineering in EECS. “What draws me most to this challenge is the work it demands: Operating as the red team puts us at the boundary between adversarial AI and real-world safety. I believe the systems we build here can have real utility in practice, not just as research artifacts but as tools that meaningfully integrate AI into real-world penetration testing.”
Each team receives $250,000 in sponsorship and monthly Amazon Web Service credits. The winning model developer team and winning red team will each receive $250,000, split among students; second-place teams will earn $100,000. In addition to Penn State, remaining teams include seven universities from the United States — University of California, Los Angeles; University of California, Santa Cruz; Carnegie Mellon University; University of Illinois Urbana-Champaign; Johns Hopkins University; The Ohio State University; and Purdue University — and two teams from Europe — Czech Technical University in the Czech Republic and NOVA School of Science and Technology in Portugal.
The Lion-0xA Team
In addition to Lee and Toufikuzzaman, the team roster includes:
- Tianrong Zhang: Co-leader, fourth-year graduate student pursuing a doctoral degree in informatics in the College of IST
- Chaewan Chun: Doctoral candidate in informatics in the College of IST
- Durva Dev: Second-year graduate student pursuing a master’s degree in computer science and engineering in EECS
- Owen Dransfield: Third-year undergraduate student pursuing a degree in cybersecurity analytics and operations in the College of IST
- Ali Al Lawati: Doctoral candidate in informatics in the College of IST
- Jason Lucas: Doctoral candidate in informatics in the College of IST
- Tanishka Mali: Graduate student pursuing a master’s degree in cybersecurity analytics and operations from the College of IST
- Jiaxi Yang: Second-year graduate student pursuing a doctoral degree in informatics from the College of IST
- Maguire Younes: Third-year undergraduate student pursuing a degree in cybersecurity analytics and operations in the College of IST
“It’s an honor to represent Penn State in the Amazon Nova AI Challenge and explore how humans and AI can code together,” Zhang said. “Our team will look beyond cybersecurity to study how people can use coding assistants more effectively. By making these tools more powerful and easier to use, we hope to open the door to more accessible software development. This work also encourages a shift toward guiding and managing AI agents, a skill that’s becoming essential. I’m excited to work with my teammates to build a practical system that explores every aspect of using coding assistants.”