Powerful new AI models are powering a growing wave of cyber attacks, experts say

New, advanced AI models are enabling a surge in autonomous cyber attacks, according to experts. Starting September 25, a hacker unleashed an army of AI agents on the internet, successfully breaching about 100 companies, including Deere, SharkNinja, and Cummins. The cost for the AI tools amounted to roughly $5,000, or about $50 per successful breach.

Jessica Ji, a security analyst, attributes this rise to recent improvements in AI models' coding skills and increased autonomy, as they can now utilize other software to complete tasks. These AI systems can operate continuously and inexpensively, unlike human attackers. Open-weight AI models, which are publicly available, add to this threat, as attackers can use them without the higher costs of closed models.

This has raised concerns among industry leaders like Anthropic CEO Dario Amodei, who has called for potential government restrictions on open models due to the risk of widespread, devastating cyber attacks. Instances of such attacks have been observed in recent months, including one targeting OpenCode, an AI coding platform, where attackers stole money from 1.4% of its users, with some losing up to $1,700.

OpenCode defended itself with its own AI agents, which ran continuously, even after OpenAI initially refused its requests. Later, OpenAI refunded affected users. Another attack on South Korean banks, attributed to Chinese state-developed AI tools, resulted in the theft of data from an estimated 68,000 individuals. South Korean officials have urged banks to strengthen their defenses against these AI-powered threats.