Cybersecurity Firm Jesta Foils Aggressive AI-Driven Attack Campaign
In a disturbing example of the growing threat of artificial intelligence (AI) attacks, a Chinese actor has been found to have intentionally weaponized an AI model to compromise over 1,200 hosts and lay the groundwork for further malicious activity. The attack, which was orchestrated by a human threat actor with clear intent, highlights the alarming potential for AI agents to be used as tools of cyber warfare.
The AI agent in question, known as DeepSeek, was employed in a proxyjacking campaign aimed at establishing a distributed network of relay infrastructure. Proxyjacking is a tactic used to gain unauthorized access to a server by exploiting vulnerabilities in its security settings. In this case, the attacker’s goal was not to steal sensitive data or encrypt systems, but rather to create an extensive network of compromised servers that could be leveraged for future attacks.
The attack was discovered and intercepted by Tel Aviv-based AI cybersecurity firm Jesta Security, which took control of the agent and tracked its activities. According to Aviv Halfon, co-founder and CEO at Jesta, the attacker’s behavior was “fully agentic” – a hallmark of autonomous AI activity. The scale of the campaign, which targeted over 1,200 hosts in addition to around 1,000 other victims, was also characteristic of an automated attack.
The investigation revealed strong indicators that the attack originated from a Chinese threat actor, including activity tied to a Beijing time zone and the inclusion of Chinese characters in payloads. The use of AI agents as tools for malicious activities raises critical questions about accountability and liability. In this case, the attacker’s intent was clear, but the fact remains that an AI model was used to orchestrate the attack.
The Five-Day Campaign
The attack unfolded over a five-day period, with the agent conducting reconnaissance and attempting to profile the target system through hundreds of short-lived SSH sessions. The activity was characterized by a distinctive pattern: the agent connected, executed a single command, disconnected, and returned after brief pauses to “think.” This behavior was consistent across the entire campaign, suggesting that the attack was indeed autonomous.
The researchers at Jesta set up a trap by seeding the environment with items that would interest an AI model but not a human. They then waited to see how the agent would respond, observing its behavior and tracking its activities. The results were chilling: the agent’s apparent objective was proxyjacking, with the goal of creating a distributed network of compromised servers.
The incident serves as a stark reminder of the growing threat posed by AI attacks. As AI models become increasingly sophisticated, they are being co-opted for malicious purposes, often with devastating consequences. The use of AI agents in cyber warfare is a trend that will only continue to grow unless addressed.
What can we learn from this attack? Firstly, it’s essential to recognize the potential for AI agents to be used as tools for cyber attacks. Secondly, organizations must invest in robust defenses against these types of threats, including AI-powered detection and mitigation systems. Finally, it’s crucial to consider the accountability implications of AI-driven attacks – who will be held responsible when an autonomous AI model is used for malicious purposes? The answers to these questions are far from clear, but one thing is certain: the threat landscape has changed forever.
Source: Dark Reading — 2026-08-03