The Fear of a One-Man-Band Weaponizing AI Agents Is Becoming Reality

A single Chinese-speaking hacker used AI agents to attempt breaches of more than 1,200 e-commerce sites across the United States, Canada, Australia and Japan between September 25 and mid-October, with roughly 10% succeeding, according to a NorForge report cited by Reuters. NorForge estimated the campaign cost about $5,000 in tokens, or roughly $50 per successful breach, and ran on Cairn, an open-source multi-agent engine on GitHub, using a mix of DeepSeek and Anthropic's Claude Code; the hacker's prompts totaled fewer than 100 Chinese characters, and the average hack took about 23 hours. Cairn's author, the user @le1xia0, said on Twitter that "Cairn is actually an innovative type of multi-agent architecture and was not designed for hacking attacks.

More than 100 websites belonging to businesses large and small were hit by a hacking collective over the last couple of weeks, stealing data, credit card numbers, and cryptocurrency. Except the collective wasn’t a group of script kiddies, but just one person and their AI agents, according to Reuters https://www.reuters.com/legal/government/how-one-hacker-used-ai-agents-few-thousand-dollars-break-into-scores-sites-2026-10-19/ , citing a recent report from cybersecurity firm NorForge https://norforge.com/blog/inside cairn . The report, published Monday, follows an unnamed Chinese-speaking hacker who allegedly outsourced almost all of their attempts to breach the lax cybersecurity of dozens of websites to AI agents—specifically, ones available through Cairn https://github.com/oritera/Cairn , an open-source AI problem-solving and state-management engine available through GitHub that has become popular among hackers. Per NorForge’s report https://norforge.com/blog/inside cairn , Cairn can coordinate AI agents and give them tasks to identify weaknesses and vulnerabilities in specific systems, and even break into those systems. Given the vibe-hacking nature of the platform, though, many hackers have left their Cairn panels exposed, allowing security researchers not only to identify them but to watch their actions, including the back-and-forth of prompt-and-response between the hackers and the AI model. One of the exposed operations that NorForge spotted involved the aforementioned Chinese-speaking hacker, who launched an elaborate campaign to hit e-commerce sites across the United States, Canada, Australia, and Japan. The hacking spree started on September 25 and, as of last week, had attempted to hit more than 1,200 sites—though not every attempt resulted in a successful breach. In fact, only about 10% got through, but that’s still a ton of breached sites. What the report found noteworthy about this AI-supported hacking campaign is just how little actual training and tuning it required. Per Reuters https://www.reuters.com/legal/government/how-one-hacker-used-ai-agents-few-thousand-dollars-break-into-scores-sites-2026-10-19/ , the hacker laid out their hacking prompts in fewer than 100 Chinese characters. The NorForge report found that some hacks were so non-technical that the prompts would be little more than “fetch credit cards” or “fetch crypto.” The agents sent out on their mission could breach sites in under a day, with the average hack taking about 23 hours to complete its task. Cairn was the harness that facilitated the hack, but the underlying model, per NorForge https://norforge.com/blog/inside cairn , was largely a mix of China’s DeepSeek and Anthropic’s Claude Code. The security researchers estimated that the widespread hacking effort that poked holes in over 1,000 sites cost about $5,000 in tokens to operate. At a 10% hit rate, that came out to about $50 per successful breach, by the company’s math. The hacks were also apparently more elaborate than standard script-kiddie attacks. One hack observed by a NorForge security researcher and recounted by Reuters https://www.reuters.com/legal/government/how-one-hacker-used-ai-agents-few-thousand-dollars-break-into-scores-sites-2026-10-19/ involved an AI agent scanning digital records to find a public-facing control panel, then crafting an exploit that eventually tricked the server into executing malicious commands. In a post on Twitter https://x.com/le1xia0/status/2102761293082223000 , the author of Cairn, a user with the handle @le1xia0 said, “Cairn is actually an innovative type of multi-agent architecture and was not designed for hacking attacks. If it is used to carry out harmful hacking attacks, that is not my intention.”