Artificial-intelligence agents were used to conduct a cyberattack campaign that successfully breached more than 100 organisations, according to an investigation by cybersecurity researcher Snorre Fagerland of NorForge reported by Reuters on Monday. The findings raise concerns that autonomous AI systems could allow cybercriminals to scale attacks while requiring comparatively little direct human intervention.
The campaign began on 25 September, with a Chinese-speaking hacker directing AI agents to probe websites operated by businesses in the United States, Canada, Australia and Japan. By late the previous week, the agents had attempted to penetrate more than 1,200 websites, with successful intrusions typically involving around 23 hours of automated probing and testing for vulnerabilities, Reuters reported.
The researcher identified the activity while scanning the internet for exposed instances of Cairn, an open-source AI problem-solving platform reportedly used by hackers. The observed campaign appeared to use a combination of DeepSeek and Anthropic’s Claude Code, although the reported use of these tools does not establish that their developers were involved in or endorsed the activity.
The financial implications could be significant. Fagerland estimated that the campaign consumed approximately $5,000 in AI tokens, or around $50 per successful breach. If comparable results can be achieved consistently, the economics could make attacks more accessible to criminal operators and increase the number of targets that a small group can pursue. The estimate is specific to the observed campaign and should not be treated as a universal cost per cyberattack.
Several companies were identified among the affected organisations. SharkNinja said the incident involved a standalone website carrying product-support information and that the threat had been contained. Engine manufacturer Cummins said it was investigating a report of a potential cyberattack on a website operated by a third-party vendor. Reuters also reported that two smaller businesses acknowledged being compromised. The extent of any unauthorised data extraction remained unclear in the initial reporting.
The incident illustrates a changing cybersecurity risk. AI agents can potentially automate repetitive tasks such as reconnaissance, vulnerability testing and the execution of multi-step instructions. That capability may increase the speed and scale of attacks, although the effectiveness of such systems still depends on the quality of their tools, instructions and access to vulnerable targets.
For companies, the implications extend beyond information-technology departments. Third-party websites, customer-support systems and external service providers can become routes through which attackers reach business operations or sensitive information.
Organisations should review internet-facing systems, enforce strong authentication, patch known vulnerabilities, monitor unusual automated activity and maintain tested incident-response procedures. Vendor security assessments are also important where external systems support customer interactions.
The next priority for investigators is to establish the full scope of the breaches, determine whether data was stolen and identify whether similar campaigns are operating elsewhere.
