As AI hacking tools that anyone can download spread, the burden on defenders appears to be growing. AI-based hacking is still in its early stages, but an analysis says open-source tools available for free are fundamentally changing the economics of cybercrime.
Axios reported on Tuesday that the U.S. Federal Bureau of Investigation warned that hackers are using the AI agent framework Artex to attack organisations around the world. Some attacks took less than 3 hours from selecting a target to gaining administrator privileges.
Amazon said hackers used Artex to steal access information for AI models and then used those resources to carry out further attacks. CrowdStrike confirmed that a hacking group used Artex over 11 days last month to target more than 200 organisations worldwide.
Palo Alto Networks said Chinese-speaking hackers used another open-source framework, Cairn, to carry out 25 attacks against at least 19 organisations between August and September. Google also said hackers used Cairn to infiltrate the U.S. aerospace sector.
AI hacking works by having AI agents draw up attack plans once hackers specify the targets and objectives. The agents then look for vulnerabilities and test intrusion paths with hacking tools. The agents share results and change tactics if they fail. Axios said they can operate around the clock without humans constantly watching.
Sam Rubin (샘 루빈), a senior vice president at Palo Alto Networks Unit 42, said, "Attacks have never been as active as they are now." He said open-weight models are nearing state-of-the-art levels and, combined with tools such as Artex and Cairn, are dealing a heavy blow to organisations.
John Hultquist (존 헐트퀴스트), chief analyst at Google Threat Intelligence Group, said incident response firms worldwide are reaching their limits and that some are turning away requests because they lack staff. He also said autonomous attacks based on AI agents may be easier to detect because they leave many suspicious traces, but for hackers seeking money, it is enough to extract key information before they are stopped.