← Back to ICOA 2026 Press Release · For Immediate Release

From “Start” to “Stop”: World’s First AI Security Olympiad Crowns Its Champions in Sydney

65 competitors from 20 countries and regions competed entirely at the command line across three frontiers of AI security — as teammate, as target, and embodied — at the inaugural International Cyber Olympiad in AI (ICOA 2026).

Sydney, Australia · 3 July 2026

SYDNEY — The International Cyber Olympiad in AI (ICOA 2026), “The World’s First AI Security Olympiad,” concluded on 2 July 2026 at the University of Technology Sydney, after six days of competition drawing 65 of the world’s most promising young cyber-AI talents from 20 countries and regions — 10 on-site and 10 remote delegations.

Hosted by the Australian STEM and Robotics Advancement Association (ASRA), ICOA 2026 was built on a single, forward-looking premise: AI is your teammate. AI is your target. AI is embodied.

A new educational paradigm — at the command line

Every second of ICOA 2026 took place inside a command-line interface (CLI) — no browser, no graphical hand-holding. As the world moves from the era of computational thinking to the era of AI-agentic thinking, the terminal is where humans and AI agents truly meet.

The organizers frame the shift simply: tomorrow’s students must do more than press Start. They must understand the underlying mechanics of how AI works deeply enough to know when — and how — to press Stop.

Two rounds that reinvent the CTF — AI4CTF + CTF4AI

Capture-the-flag (CTF) is a discipline of its own, spanning five technical domains and traditionally requiring a five-person team. ICOA’s headline innovation collapses that team into one person — with AI as the fifth teammate.

• AI4CTF — “AI as your teammate.” For the first time, a single competitor takes on all five CTF disciplines alone, using AI as a genuine collaborator rather than a search box. To guarantee a level playing field, every competitor worked with the same model — Google’s Gemini 2.5 Flash Lite — balancing useful assistance with competitive integrity. The intensity was real: across a five-hour endurance round, some competitors were still actively working the AI right up to the final minutes.
• CTF4AI — “AI as the target.” Here the tool becomes the objective: AI itself is the protected system to be probed, evaluated and broken. Operating at the frontier of AI security, the challenges were co-designed by two members of the MEA Blackhack 2025 world-champion team, with a carefully engineered difficulty gradient so that high-school competitors — most arriving without a formal cybersecurity background — could climb from first principles to genuine exploitation.

The combined results of these two rounds decided ICOA 2026’s Gold, Silver and Bronze medals.

The Embodied-AI Arena — a world first (CTF4EAI Pioneer Round)

Reflecting the rise of embodied intelligence, ICOA 2026 introduced a standalone CTF4EAI Pioneer Round — the “Embodied-AI Arena.” Founding partner DeepCybo supplied the embodied-AI (robotic) model, and the server generated live challenge scenarios inside a MuJoCo physics-simulation environment. Competitors used natural language to probe the robot’s policy for unsafe behaviour — taking AI security out of the browser and into the physical world.

15 competitors advanced from the preliminary round to the final, where a four-task, 25-minute tie-break decided the podium:

The champions

ICOA 2026 awarded 36 medals and honours — 6 Gold, 11 Silver, 17 Bronze, and 2 Honorable Mentions — each certified after a rigorous, multi-dimensional integrity review.

Team Medal Table (Top 5):

Gold Medalists: Oleg Uteshev (overall champion, Central University) · Roman Mielamud (Ukraine) · Gleb Filatov (Central University) · Timofei Gulenok (Central University) · Georgii Kalita (Central University) · Hao Hui Liu (Macau, SAR, China).

Honorable Mentions: Funny-Chryssie Irishura (Burundi) · Romina Phoutharaksa Douangmala (Laos).

Backed by an international scientific committee

ICOA 2026 was overseen by an International Scientific Committee of leading academics, ensuring the rigor of every challenge and result. It was chaired by Professor Paul Haskell-Dowland, Professor of Cyber Security Practice at Edith Cowan University, and included Distinguished Professor Fang Chen of the University of Technology Sydney, who called the Olympiad “a really exciting opportunity” for the next generation of AI-security talent.

A founding moment

Delegates from the competing nations signed The Founding Charter of ICOA, formally establishing the world’s first AI-security olympiad as a lasting international institution.

Momentum into 2027

The inaugural edition has already sparked worldwide demand: 20 additional countries have been accredited for ICOA 2027, positioning the Olympiad to become the definitive international proving ground for the AI-security generation.

---