@MaxForAI: In Anthropic's report, another particularly intriguing case is: Two undergraduates who are already capable of participa…

In Anthropic's report, another particularly intriguing case is: Two undergraduates who are already capable of participating in a 24-hour nonstop Agent network attack system. Anthropic tracked a possible Chinese-language cyber operations team located in Changsha (case number GTG-10007), in which two operators were confirmed to be undergraduates from the College of Computer and Communication Engineering at a certain university in Hunan. One of them had even interned at Sangfor Technologies and is currently interviewing for an offensive cyber operations position at Qi An Xin. This group uses Claude as the engineering and orchestration layer for the entire attack system, automatically reverse-engineering firmware, hunting for vulnerabilities, generating exploits, testing them, and continuing to iterate after failures. One of their automated workflows produced over a dozen potential 0-day vulnerabilities in just one month. The so-called 0-day vulnerabilities refer to security flaws that have not yet been publicly disclosed, for which there are no official patches, and which the software or hardware vendors are typically unaware of. When security maintenance personnel encounter this type of vulnerability, the time for fixes and responses is zero days (meaning they cannot be defended against). At the same time, they also ran 13 resident AI Agents, continuously collecting content from U.S. military, government websites, procurement information, and social media, then automatically compiling it into intelligence. The most fascinating part is that these Agents continue to work even when the humans are offline. What used to require a security team to invest continuously is now starting to turn into a 24-hour automated production line. This case already illustrates quite clearly how Agents amplify the capability boundaries for individuals and small teams.