Autonomous AI Agents Fueling New Wave of Cyberattacks
AI-powered agents are executing autonomous breaches at a fraction of the cost of human attackers.
1Autonomous Breach Campaign
Starting September 25, an autonomous AI agent campaign targeted over 1,200 websites. The operation successfully breached 100 companies, including Deere, SharkNinja, and Cummins. The hacker required no manual intervention after launching the agents, which operated continuously to identify and exploit vulnerabilities across various corporate digital infrastructures without further human input.
2Shift in Attack Economics
The campaign utilized DeepSeek and Anthropic’s Claude Code models, costing approximately $5,000 in total tokens. This expenditure resulted in a cost of roughly $50 per successful company breach. Experts note that this fundamentally changes the economics of exploitation, as AI agents can operate 24/7 at a fraction of human costs.
3Enhanced Agentic Capabilities
AI models have evolved significantly over the past year, gaining improved coding skills and the ability to analyze large code volumes. Beyond text generation, these agents are now 'agentic,' meaning they can actively use external software to complete complex tasks. This transition enables systems to perform sustained, autonomous cyberattacks without stopping.
4Open vs. Closed Models
Debate persists regarding the risks of open-weight models, which are publicly downloadable and customizable. While closed models have strict security guardrails, open models offer powerful, low-cost tools for attackers. Some industry leaders suggest government restrictions on distribution, while others argue that open models are essential for defenders to build effective countermeasures.
5Defensive AI Limitations
Hugging Face and OpenCode both utilized AI agents to defend their systems during recent attacks. However, they faced challenges when closed-model providers like OpenAI restricted access due to security guardrails. OpenCode eventually gained access to advanced tools through a verified cybersecurity program after experiencing significant frustration during the active 48-hour attack.
6Financial Impact on Users
During the OpenCode incident, attackers compromised user accounts and stole funds, with some individuals losing as much as $1,700. In a notable response, OpenAI refunded the affected users. This incident highlights the direct financial risks posed to end-users when AI-driven platforms are successfully breached by autonomous agentic hacking tools.
7Large-Scale Data Theft
A severe attack this month targeted South Korean banks, resulting in the theft of data from approximately 68,000 individuals. The attackers, likely based in China, employed a combination of the Artex AI tool and Anthropic’s Claude Code. South Korean officials have since urged financial institutions to significantly harden their existing security defenses.
8Future Security Outlook
Cybersecurity experts anticipate a global rise in AI-powered agentic attacks. As models become more capable and accessible, the barrier to entry for sophisticated exploitation continues to drop. Organizations must now prepare for persistent, non-stop threats that operate autonomously, necessitating a shift toward more resilient and adaptive AI-driven defensive security architectures.
Original article: Fortune — Autonomous AI Agents Fueling New Wave of Cyberattacks