Sales November 2024 Ten top university teams from all over the world have gathered in the initial Amazon Nova AI challenge: trusted AI tracks, focused on strengthening the security of AI cod assistants and developing new automated methods to ‘red-team’ and testing them. After months of intense competition, eight teams have moved on to the final, demonstrating unique innovation to secure AI-hugged code generation. The final will take place 26-27. June, with Judes Conventing in Santa Clara, California, while teams participate in a remote in a tournament style competition designed to push the boundaries of safe AI-Assisted software development. In each tournament, attack and defend teams match each other, and in each match the striker engages a limited number of conversations with the defender to try to request malicious code, vulnerable code or help with malicious cyber activity. In addition to their success in defending, defense models are also regained for their benefit to supporting coding tasks. Attacking systems are evaluated by attack success and the diversity of their attacks. In the final, in addition to the final tournament model defense and attacks, there will also be evaluated by expert human red-teamers.
The finalists
The finalists were chosen on the basis of tournament services, research articles and presentations of their innovations.
Defense team (model developers)
Attack teams (security testers)
“Sales November, all teams have developed increasingly innovative ways to make AI-Assisted Coding more secure and reliable,” said Michael Johnston, head of Amazon Nova Ai Challenge. “The quality of the work has been unusual, making the finalist selection very competitive.”
Defense Teams is working to build robust security features in Codegeneract models, while the striker teams develop sophisticated techniques to test thise models and identify potential vulnerabilities. Together they help shape the future of safe AI development.
The final format
In the final, teams will compete for remote in an offline tournament evaluated by a panel of Judes from Amazon’s artificial general intelligence (AG) -Team, Amazon Security, AWS Responsible AI and AWS Q for developers. The final tests the teams’ solutions to scenarios in the real world in a controlled competitive environment.
“The challenge sits at the intersection of AI capacity and security, two critical areas for the responsible promotion of generative AI,” Dr. Gang Wang, one of the Faculty Advisors for the UIUC team. “Our students have well developed to develop new approaches that improved security without compromising the user experience.”
After the final, all teams in Seattle gather on 22-24. July to Amazon Nova Ai Challenge Summit, where the winners will be content and team will present and share their research results.
Promoting the field
The challenge brings together Amazon’s AI expertise with top academic talent to speed up safe innovation in generative AI. Research produced through competition aims to contribute to the development of safer, more reliable AI systems for everyone.
“What makes this challenge particularly valuable is that it is combined technical innovation with the real world application,” Dr. Xiangyu Zhang from Purdue University. “Our students don’t just compete, they help solve AI security challenges in the real world.”
Amazon Nova Ai Challenge is part of Amazon’s broader commitment to responsible AI development and academic collaboration. For more information and updates, visit Amazon.science/nova-i-Ongge.