Agents

International hacking conference draws 405 security researchers from 37 countries to discuss AI's impact on vulnerability discovery

November 19, 2026 · 3 min read

woman in white and red polka dot long sleeve shirt
Jeswin Thomas / Unsplash

The Power of Community 2026 conference, an international hacking and security event, took place November 12 and 13 in Seoul. According to organizer POC Security, the event attracted 405 participants from 37 countries, including speakers, attendees, and sponsor representatives. The two-day program featured 14 technical presentations and one special panel discussion, with on-site registration closing after capacity was reached.

POC began in 2006 as South Korea's first international hacking conference and has grown into a technical forum where researchers share how they discover and verify vulnerabilities. This year's presentations covered security breaches across different systems including web browsers, smartphones, Windows kernels, virtual machines, and containers. A recurring theme throughout the program was how artificial intelligence is changing the speed of vulnerability discovery and patching, requiring researchers and defenders to reconsider their standards.

On the first day, a Seoul National University researcher analyzed the V8 JavaScript engine sandbox used in Chrome, examining why the mechanism designed to prevent code executed in browsers from accessing unauthorized areas failed in multiple cases. Researchers from Taiwan-based security firm DEVCORE demonstrated a Safari 26.5 vulnerability during a live presentation, showing how multiple browser flaws could combine to allow access to operating system files beyond the browser's isolation boundary. The presentation illustrated how browser defects could expand their impact into the operating system domain.

Other presentations examined iOS security through research on vulnerabilities named Momentarius and Coruna, Windows system component management issues involving TrustedInstaller, and remote code execution vulnerabilities in OpenAI Codex Desktop. On the second day, researchers analyzed security boundaries in QNX hypervisors separating virtual machines from hosts, and cases where isolation boundaries could break in Linux containers using NVIDIA GPUs. A KAIST hacking lab shared findings from using artificial intelligence over one month to discover 45 vulnerabilities.

A special panel titled Hacking in the Age of AI featured Professor Kim Tae-soo and Microsoft researchers. Kim's team Atlanta won the 2025 Defense Advanced Research Projects Agency artificial intelligence cyber challenge, and team members subsequently joined Microsoft to continue artificial intelligence security research. Organizer Jung Jin-wook noted that the panel's significance lay in having researchers who directly experienced artificial intelligence finding and fixing vulnerabilities share their experience firsthand.

Jung identified the 2016 DARPA Cyber Grand Challenge, held in conjunction with DEF CON, as a turning point in vulnerability research automation. He explained that a decade later, artificial intelligence can now analyze code in hours that humans previously spent weeks examining, changing not only the time and cost of vulnerability discovery but also the time allowed for patch deployment and standards for evaluating attack novelty. Jung emphasized that the field must redefine what constitutes good research and sufficient defense, requiring both forums for sharing research and trial-and-error experiences alongside institutional changes. He raised concerns that as artificial intelligence analysis accelerates, humans may be relegated to merely confirming and approving results, arguing that discussions must begin now about what to protect and which decisions should remain human responsibilities.