On August eight in Las Vegas, the air was heavy with a tense silence. It was the long-awaited second the winner of the world’s largest safety know-how competitors — the AI Cyber Challenge (AIxCC) — can be revealed.
All eyes had been mounted on the display screen, set to announce the champion. Then, when the Team Atlanta brand appeared, the silence erupted into thunderous cheers and applause. It was the end result of two years of relentless effort and willpower, which had resulted within the workforce rising above world-class rivals to win the first-place prize of $four million.
▲ Team Atlanta wins first place on the AI Cyber Challenge.
To take readers behind the scenes of their inspiring journey, Samsung Newsroom sat down with Taesoo Kim, Corporate Vice President, and Joonun Jang from Samsung Research’s Security & Privacy Team, together with Yunjae Choi from the AI Productivity Team.
▲ (From left) Taesoo Kim, Joonun Jang and Yunjae Choi from Samsung Research
Team Atlanta: A 40-Strong Global Alliance of Top Researchers, Engineers and Hackers
The AI Cyber Challenge is a worldwide safety know-how competitors hosted by the U.S. Defense Advanced Research Projects Agency (DARPA). Sponsored by main tech firms together with Google, Microsoft and OpenAI, the newest version of the occasion featured a complete prize pool of $22.5 million. At this occasion, Team Atlanta — a coalition of greater than 40 world-class safety researchers from Samsung Research, the Georgia Institute of Technology, KAIST, POSTECH and different main establishments — got here collectively beneath the management of Taesoo Kim, who additionally performed a pivotal function in assembling the workforce.
▲ Taesoo Kim explains what impressed the workforce to tackle the problem.
Also a professor on the Georgia Institute of Technology, Kim reached out to a number of of his former Ph.D. college students after first studying concerning the AI Cyber Challenge and inspired them to hitch. They quickly grew to become the core members of Team Atlanta.
Under Kim’s management, Team Atlanta introduced collectively expertise from throughout Korea and each the East and West Coasts of the United States. “Because our members came from such diverse locations, we structured our collaboration framework around each member’s base,” he recalled. “Some of our colleagues in Korea even carved out time from their busy schedules to travel to the Georgia Institute of Technology so we could work side by side.”
How Team Atlanta Outpaced the Competition
The AI Cyber Challenge is a contest designed to check the flexibility to establish and remediate vulnerabilities in large-scale software program methods. Its final goal is to speed up the event of AI-powered safety applied sciences able to defending vital infrastructure corresponding to transportation, vitality and healthcare. In the ultimate spherical, individuals had been tasked with constructing an AI-driven “Cyber Reasoning System (CRS)” able to autonomously detecting software program vulnerabilities and producing safety patches.
Scoring was primarily primarily based on two elements: vulnerability discovery and program restore. Teams earned discovery factors for precisely figuring out weaknesses within the given code, and restore factors for efficiently patching them. In brief, a workforce’s rating mirrored how shortly its system may detect vulnerabilities within the organizer’s supply code, what number of it may detect and the way precisely it may repair them.
▲ The first-place trophy and Team Atlanta’s profitable scorecard
Team Atlanta clinched the championship with a decisive lead, uncovering extra vulnerabilities and patching them with higher accuracy than their rivals.
Jang, who oversaw the event of a big language mannequin (LLM) agent for detecting vulnerabilities in Java packages, defined that the agent helped Team Atlanta’s system detect extra vulnerabilities throughout the competitors. “Team members every developed their brokers in numerous methods,…
