GCSA Security AI Agent Achieves 91.3% in CyberGym Evaluation

The GCSA's security AI agent recorded a success rate of 91.3% in the CyberGym evaluation. This agent is based on the Grok 4.5 and 4.6 models, and CyberGym is an AI security agent evaluation system that includes 1,507 actual vulnerabilities from 188 software projects. The focus of the evaluation is not on code comprehension but on the ability to reproduce actual vulnerabilities, with the success of the AI agent determined through vulnerability descriptions, code analysis, and attack path inference. GCSA describes this result as a performance of the AI security agent and has presented cases where AI discovered zero-day vulnerabilities and incomplete security patches. However, the results on the CyberGym leaderboard may vary depending on team submissions, and the score differences may not reflect actual capabilities. The 91.3% result of the GCSA agent indicates high performance, but further verification is needed to assess the impact of this score on service security quality.

This content is provided for general informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online promotions, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets. Crypto assets are highly volatile and may result in loss. The availability of WEEX services, products, and related events may vary by region. You are responsible for ensuring that your participation is in accordance with applicable local laws and regulations.