Hacking with AI armies: the new paradigm of cybercrime

Automation pushed to the limit: A new threat paradigm

The global cybersecurity ecosystem is facing an unprecedented crossroads. Recent investigations by the specialized firm Norforge have uncovered a massive intrusion campaign where an attacker managed to breach over 100 websites using a network of autonomous artificial intelligence agents. Unlike traditional hacking tactics, which usually require constant human interaction for vulnerability exploitation or lateral movement, this operation was executed with minimal intervention. Industrial and consumer giants like John Deere, SharkNinja, and Cummins have been identified by outlets such as Reuters and The Verge as the primary targets of this assault, proving that not even corporations with robust cybersecurity budgets are immune to high-level automation.

This event is not an isolated incident, but the culmination of a trend we have been observing at TheVortiq: the democratization of offensive capability through autonomous agents. Historically, the complexity of a large-scale cyberattack limited the actors capable of orchestrating it to nation-states or highly coordinated organized crime groups. The arrival of tools capable of reasoning and executing complex tasks autonomously has drastically lowered the barrier to entry, allowing a single individual, with the right technical resources, to multiply their operational footprint exponentially.

The tool behind the attack: Cairn and the democratization of the exploit

The central attack vector in this operation was Cairn, an open-source artificial intelligence platform originally conceived as a general-purpose problem-solving engine. The architecture of Cairn allows users to define an initial state and a final goal, delegating to the AI the ability to document its progress and adjust its tactics in real-time. What its developers designed as a productivity tool, attackers have transformed into a highly efficient attack infrastructure.

According to data collected by Norforge, the attacker managed to orchestrate multiple simultaneous intrusions in the United States, Canada, Australia, and Japan using a series of extremely brief instructions, composed of fewer than 100 characters in Chinese. This level of abstraction in the command is alarming: the system not only executes but interprets the environment to find the path of least resistance. Technical reports indicate that, in the analyzed instances, the time elapsed from the agent's activation to the successful breach was just 23 hours. Furthermore, Norforge has identified more than 20 exposed instances of Cairn that exhibit behaviors indicative of malicious activity, suggesting that the incident involving John Deere and other companies is only the tip of the iceberg of a network of agents operating in the shadows.

Impact and corporate response: The fragility of the digital supply chain

The reactions of those affected offer an X-ray of the current state of risk management. John Deere has communicated that there is no evidence of access to sensitive information, while SharkNinja clarified that the incident was limited to a product support portal. For its part, Cummins is in the process of investigation, focusing specifically on a website managed by an external provider. This segmentation of responses underscores an uncomfortable truth: the digital supply chain and third-party services have become the most vulnerable links in modern corporate infrastructure.

If we compare this event with historical breaches like the SolarWinds attack, the difference lies in speed. While in the past attackers needed weeks or months to infiltrate and deepen their reach into networks, AI allows for vulnerability scanning in real-time and at a massive scale. Companies are not being attacked by people, but by autonomous workflows that can test thousands of configurations in minutes.

Implications for the future of work and security: Towards a Zero Trust model

This incident acts as a catalyst for rethinking security architectures. An attacker's ability to orchestrate complex intrusions via minimal commands suggests that the future of cybercrime will be, effectively, silent, fast, and highly scalable. We are entering an era where manual defense is mathematically unfeasible.

The response of organizations must necessarily transition towards a Zero Trust architecture reinforced by defensive automation. In this model, access is not granted by default to any actor or process, and network segmentation must be granular enough to contain an AI agent that has managed to infiltrate. The lesson of Cairn is clear: AI has leveled the playing field, and companies that do not adopt defense systems based on the same technology as their attackers (defensive AI, real-time anomalous behavior monitoring) will be operating at a competitive and operational disadvantage.

It is important to note that, although the potential for autonomous agents to do harm is evident, the security community is still debating to what extent these systems can be contained in test environments. The risk of "rogue" agents escaping their controlled environments, as has been observed recently in security research, adds an additional layer of uncertainty that boards of directors must begin to consider as a priority operational risk for the coming years.