AI OCT 14, 2026

NICT and MITRE to Host AI Agent Security Competition in Washington D.C.

Japan's national ICT research institute and U.S. nonprofit MITRE will run a 24-hour red-team competition testing vulnerabilities in email AI agents, December 2026 in Virginia.

TOKYO — Japan’s National Institute of Information and Communications Technology (NICT) and U.S.-based MITRE Corporation will co-host a 24-hour competition testing the security of email AI agents, December 16–17, 2026, at George Mason University’s Arlington, Virginia campus.

The “AI Breach Challenge 2026: Email Agent Security” will task teams of two to four participants with probing vulnerabilities in AI agents designed to handle email and workplace tools. Competitors will attempt to force unintended actions, access protected data, and expose failure modes in a controlled environment. Registration and rules will be posted at create.nict.go.jp/ai-breach-challenge-2026; fully remote participation is not permitted.

NICT launched its AI Security Research Center (CREATE) in February 2025; MITRE maintains ATLAS, a knowledge base cataloging adversarial tactics against AI systems. The competition serves as a research exercise: findings will inform future agent defense and evaluation techniques. No prize pool was disclosed.

Why it matters: AI agents that execute tasks autonomously — booking meetings, drafting replies, querying databases — are entering enterprise workflows faster than security frameworks can adapt. Real-world attack scenarios remain poorly understood. The organizers plan recurring challenges covering other agent use cases to build a Japan-U.S. research baseline as deployment accelerates.