Cybersecurity Experts Sound Alarm as Autonomous AI-Driven Attacks Emerge
A sophisticated threat actor has been using a cutting-edge artificial intelligence (AI) model called DeepSeek and an open-source agent called Hermes to launch autonomous cyberattacks on exposed servers. The attacks, which were discovered by Palo Alto Networks’ Unit 42 researchers, demonstrate the alarming capabilities of AI-driven malware in identifying, evaluating, and exploiting vulnerable systems.
The threat actor, who goes by the aliases “knaithe” and “KnYuan,” claims to be a “binary security researcher” from China. While their attacks did not successfully compromise any servers, the campaign reveals an end-to-end autonomous offensive capability that can rapidly scan for vulnerabilities, download exploit code, and execute attacks in minutes what would normally take hours.
At the heart of this operation is DeepSeek, an AI model used as the reasoning engine behind Hermes Agent. This open-source framework allows the agent to interact with operating system terminals, run commands, and connect to the internet. What’s particularly concerning is that Hermes was configured to operate in “Yolo” mode, which enables it to execute commands without human permission.
The researchers recovered a session from May 2026 where the operator provided only an initial task, after which the agent conducted the remaining activity autonomously. The agent first targeted Langflow servers vulnerable to CVE-2026-33017, downloading a public proof-of-concept exploit and scanning over 84 exposed instances for vulnerable configurations.
However, when it determined that these targets couldn’t be exploited, the agent searched for other potential vulnerabilities. It analyzed multiple public exploit repositories before selecting the n8n workflow automation platform as its next target. The agent downloaded an exploit that chained CVE-2026-21858 and CVE-2025-68613, identified vulnerable servers running outdated versions of the platform, and checked for unauthenticated file-upload forms required to complete the attack.
The attempts ultimately failed due to the presence of authentication requirements, but this exercise illustrates the speed and efficiency with which AI-driven malware can operate. Unit 42 notes that what would normally take hundreds of hours of manual targeting analysis was accomplished in mere minutes by the autonomous agent.
This campaign is significant not only because it showcases an advanced AI workflow but also because it highlights the potential for widespread exploitation. The threat actor conducted manual attacks against over 460 systems using vulnerabilities affecting various products, including Citrix NetScaler, Apache Tomcat, and Windows IKE VPN.
While this incident serves as a warning about the growing threat of autonomous AI-driven attacks, there are steps that organizations can take to mitigate this risk. By regularly updating software, implementing robust security protocols, and monitoring for suspicious activity, businesses can reduce their vulnerability to these types of attacks.
As AI technology continues to evolve, cybersecurity professionals must remain vigilant in identifying and responding to emerging threats. The exposed campaign serves as a reminder that the cat-and-mouse game between attackers and defenders is becoming increasingly complex, with AI playing an ever-larger role on both sides.
Source: Bleeping Computer — 2026-07-31