A recent study by Stanford University revealed that an artificial intelligence (AI) agent named ARTEMIS surpassed most human hackers in identifying vulnerabilities on a university website after just 16 hours of scanning. Amid rising cyber threats from state-sponsored hackers and militant groups using large language models (LLMs), ARTEMIS showcased a high level of technical sophistication, ranking second among ten human testers. The AI discovered nine vulnerabilities and submitted 82% of its findings as valid. Costing only $18 per hour compared to $60 for professional penetration testers, ARTEMIS’s efficiency highlights its potential in cybersecurity. While it generated “sub-agents” to assess vulnerabilities simultaneously, ARTEMIS still missed some issues identified by humans. The study noted that existing AI cybersecurity tools from OpenAI and Anthropic lacked the necessary expertise, further emphasizing ARTEMIS’s unique capabilities. This research indicates a shift in the cybersecurity landscape where AI agents may soon play a critical role in threat detection and response.
Source link
Share
Read more