AI-Powered Cyberattacks: Autonomous Execution Through Prompts and Credentials

Most cyberattacks now begin with AI prompts, where attackers provide AI agents with URLs and access credentials, often obtained from stealer logs purchased from initial access brokers on the dark web, and the AI executes the attack autonomously. The speaker describes a recent rapid response effort at work where they conducted complete security research, including proving a new vulnerability, uncovering additional ones, and adding telemetry for network data collection, entirely through conversational interaction with AI without ever opening a virtual machine. This represents a fundamental shift in how security research and offensive operations are conducted, demonstrating how accessible and streamlined these processes have become through AI assistance.