Amazon Announces 2026 Amazon Nova AI Challenge: Trusted Software Agents Track (www.amazon.science)

Amazon today opened the 2026 Amazon Nova AI Challenge — Trusted Software Agents track, a university competition that asks student teams to build multi-step, agentic systems that plan, execute, and validate changes across real codebases while demonstrably improving secure-coding performance. Ten teams will be selected as either developer teams (engineering defenses and reliability into agentic coding systems) or red teams (actively probing for failures and security weaknesses). Evaluations intentionally pair utility and safety: judges will measure how well agents complete complex feature development and refactoring tasks alongside adherence to guardrails, exploit resistance, and measurable gains in secure-coding outcomes. Applications open November 10, 2025 via YouNoodle and the program runs across the 2025–26 academic year with benchmarks, live tournaments, and shared resources. This track is significant because it shifts the focus from single-step code generation to real-world agentic workflows that can affect product quality and system integrity at scale. By requiring simultaneous progress on task effectiveness and safety, Amazon is promoting research and engineering practices—like red-team-driven evaluation, integrated verification and testing steps, and joint utility-safety metrics—that are crucial as AI systems gain autonomy over code. The challenge signals industry interest in standardized, real-world benchmarks for trustworthy software agents and will likely accelerate techniques for secure planning, validation, and deployment in agentic software engineering.