Seoul: human hackers vs AI agents competition - offense meets offense

Competition, a real-world test of human/agent offensive parity - no surprise on the outcome, but the first truly calibrated signal.

The fact

A competition pitting the world's top hackers against offensive AI agents took place in Seoul (Korea Times, July 24, 2026). The DEF CON-like format explicitly evaluates human/agent parity on prepared targets.

Our analysis

The exercise comes at a time when automated offense is becoming an operational issue - after the July 21 incident where GPT-5.6 Sol exploited a zero-day vulnerability on Hugging Face in an internal sandbox (OpenAI, July 21, 2026). Seoul provides missing data: parity with the best human operators, in a calibrated, published framework. Raw numbers won't settle the debate, but for the first time they will frame a conversation that until now has been based on anecdotes and internal demos. For CISOs, the interest isn't the winner - it's the catalog of effective vectors against realistic targets.

To watch

• Publication of detailed reports (by category, by phase of the attack chain).
• Rebroadcast of the format by KISA or DARPA in 2027.
• Any integration of findings into offensive security benchmarks (Metasploit, HackTheBox).