AI Breach
Challenge 2026
Email Agent Security
A controlled adversarial exercise testing the security boundaries of autonomous AI email agents operating within enterprise communication environments.
01 / About
Competition Overview
AI Breach Challenge 2026: Email Agent Security is an adversarial competition co-organized by the National Institute of Information and Communications Technology (NICT) and MITRE. It examines how autonomous AI email agents — systems that use large language models to read, compose, and act on email — can be manipulated, deceived, or exploited when operating within enterprise communication environments.
Teams of 2–4 participants probe a realistic deployment scenario in which an AI agent holds delegated access to email, calendar, and contacts within a controlled competition environment. The competition aims to generate structured findings that inform safety-by-design principles for agentic AI systems and contribute to open research in adversarial AI evaluation.
02 / Competition
Competition Environment
The AI agent under test operates with delegated access to the following systems within the controlled competition environment. Participants attempt to exploit the agent through these interfaces.
The AI agent can read, compose, and send messages on behalf of the user within the competition environment.
The AI agent can view and manage calendar events and meeting schedules within the competition environment.
The AI agent can access and interact with contact information within the competition environment.
03 / Rules
Rules & Guidelines
The following preliminary rules outline the main expectations for participants. Detailed competition rules will be provided before the event.
1. Stay within the intended scope of the competition.
Focus your activities on the systems, information, and functions provided for the competition. Please avoid accessing, attacking, disrupting, or interfering with other systems, infrastructure, participants, or teams.
2. Participants should lead the work.
AI tools and other technologies are welcome to assist, but registered team members should meaningfully conduct and control their competition activities.
3. Compete fairly and keep competition information within your team.
During the competition, keep solutions, attack methods, results, evidence, and other competition materials within your registered team.
4. Follow the competition and submission process.
Follow task instructions and use the designated systems for submissions. If you encounter unintended access, vulnerabilities, or technical issues, please report them to competition staff rather than exploiting them.
5. Help us maintain a fair and safe competition.
Please follow organizer instructions and any rule clarifications or updates. If a rule issue arises, organizers will determine an appropriate response, which may include adjusting submissions or scores, suspension, or disqualification. Organizer decisions on rule interpretation and enforcement are final.
04 / Schedule
Schedule
December 16-17, 2026
George Mason University, Mason Square, Arlington, VA
Competition: 12:00 PM
Day 1 to 12:00 PM Day 2.
On-site check-in and participant registration at the venue.
Welcome address, rules overview, and competition environment walkthrough.
Competition begins. All teams start simultaneously.
The on-site session closes for the day. Competition continues remotely overnight.
Participants return to the venue for the final on-site session.
Competition ends. All team submissions are finalized.
05 / Participation
Who Should Participate
AI Breach Challenge 2026 welcomes teams of 2–4 members from a range of backgrounds. Whether you are a student, early-career participant, researcher, cybersecurity professional, or red team practitioner, you are encouraged to apply.
Students & Early-Career Participants
Undergraduate and graduate students, and early-career professionals with an interest in AI and cybersecurity looking to gain hands-on experience in a real-world adversarial AI setting.
Researchers
Researchers with experience in adversarial machine learning, AI safety, or related fields seeking empirical grounding through hands-on adversarial evaluation of AI systems.
Cybersecurity Professionals & Red Teamers
Cybersecurity professionals and red team practitioners with experience in offensive security who want to apply their skills to AI-mediated attack surfaces.
06 / Venue
Venue
Access
George Mason University’s Mason Square campus in Arlington, VA is accessible from three major airports: DCA (Reagan National), IAD (Dulles), and BWI (Baltimore/Washington).
Competition Format
The competition is primarily on-site. After the Day 1 session ends at 6:00 PM, the competition continues remotely overnight. Participants return to the venue on Day 2. Fully remote participation is not planned.
07 / FAQ
Frequently Asked Questions
How do I register?
To participate, please complete and submit the registration form available on this website. Participants may register either as a pre-formed team or as an individual participant.
Who can participate, and how are teams formed?
There is no age restriction, and prior experience in cybersecurity is not required. Participants may register as a pre-formed team of two to four members or individually. If you register individually, the organizers will assign you to a team before the competition. Further information about team assignments will be provided after registration.
When is the registration deadline?
The registration deadline has not yet been set. Registration may close once capacity is reached, so we encourage you to register early. Any updates will be announced on this website.
What does the competition environment include?
Participants will operate within a controlled competition environment that includes an email server, a calendar application, and a contacts system. AI agents under test will be given access to designated systems within this environment. Further details about the competition environment will be announced closer to the event.
What is the competition objective?
All participating teams will share the same objective: to identify and demonstrate potential vulnerabilities in an AI agent operating within the competition environment. Detailed challenge tasks, rules, and evaluation criteria will be announced before the competition.
What tools will be available to participants?
Participants will be provided with access to the systems and resources necessary to take part in the competition. Further information about the tools and technical environment, including whether participants may use their own tools, will be provided once the competition rules have been finalized.
What is the competition format?
AI Breach Challenge 2026 will primarily be an on-site event at George Mason University, Mason Square, Arlington, VA. Participants will begin the competition on-site and continue competing remotely overnight after the Day 1 on-site session concludes at 6:00 p.m. On Day 2, participants will return to the venue, and the competition will conclude at 12:00 p.m. Fully remote participation is not currently planned.
Is there a participation fee?
There is no fee to participate in the competition. However, participants are responsible for arranging and covering their own transportation and accommodation. The organizers will not provide lodging or reimburse travel or accommodation expenses.
What should I bring to the competition?
Participants should bring their own laptop, charger, and any other accessories needed to use their computer throughout the competition. Information about identification or other check-in requirements will be provided to registered participants before the event.
What happens after I register?
Registered participants will receive further information by email, including participation details, team assignments for individual registrants, competition rules, and event-day instructions.
How can I contact the organizers?
If you have questions about registration or the competition, please contact the organizers using the inquiry form available on this website.
Will the competition results or findings be published?
NICT and MITRE may publish a post-competition report summarizing the overall results, key findings, and techniques identified through the competition. Any publicly released information will be handled appropriately, taking into account security considerations and the treatment of team-specific information. Further details regarding publication will be provided before the competition.
08 / Organizers
Organizers
Japan's flagship R&D institute for information and communications technology. NICT conducts cybersecurity research, international standardization activities, and operates critical national network security infrastructure.
A not-for-profit organization managing federally funded research and development centers. MITRE developed ATLAS (Adversarial Threat Landscape for Artificial-Intelligence Systems), a knowledge base of adversarial tactics, techniques, and case studies targeting AI and machine learning systems.