Big tech companies continue to drive the debate on Artificial General Intelligence (AGI), although doubts persist about its actual proximity. AGI is defined as AI capable of matching or surpassing human abilities in any field of knowledge. While it is still unclear if this goal will be achieved, there are areas where AI is beginning to outperform humans, such as hacking. A recent experiment conducted by Stanford University, reported by the Wall Street Journal, illustrates this advancement. Researchers developed a hacking bot named Artemis, designed to analyze networks and detect vulnerabilities. Artemis was tested on the university's engineering network, competing against ten professional pentesters, experts in simulating attacks to identify and fix security flaws. The bot had an emergency shutdown mechanism, and the pentesters were instructed to look for vulnerabilities without compromising the network. The results surprised the researchers themselves: Artemis outperformed nine of the ten human pentesters, identifying flaws faster and at a significantly lower cost. While a pentester can charge between $2,000 and $2,500 per day, Artemis's operational cost is around $60 per hour. However, Artemis is not infallible. Approximately 18% of its reports were false positives, and it overlooked an obvious flaw that humans detected immediately. Nevertheless, the bot also identified a vulnerability that no human had seen, as it was on a website incompatible with the usual browsers of the pentesters. Artemis, not relying on browsers, was able to analyze the website and discover the flaw. The use of AI in hacking is not new. Cybercriminals have been using artificial intelligence to create more sophisticated malware for some time. Recently, Anthropic detected that a group of Chinese hackers was using Claude Code for a spying campaign, where the AI managed the entire attack cycle. But AI can also be an ally in defense. Experiments like Stanford's demonstrate that artificial intelligence can help identify and correct vulnerabilities, improving system security. The challenge now is to define the role of human professionals, such as pentesters, in an environment where AI can perform their work more efficiently and economically. Source: xataka.com