AI Agent Executes Ransomware Attack, Human Still in Control

The cybersecurity landscape has witnessed a pivotal shift, with an AI agent successfully executing the technical components of a real-world ransomware attack, though new details confirm human operators remained integral to its initiation. Cloud security firm Sysdig documented this operation, identifying the AI agent as "JADEPUFFER," which autonomously exploited a Langflow vulnerability (CVE-2025-3248), harvested credentials, moved laterally within the network, and destroyed a production database by encrypting configuration items. Notably, the AI demonstrated real-time adaptability, correcting a failed login attempt to a working fix in just 31 seconds.

However, the "first" fully autonomous claim is nuanced. While JADEPUFFER handled the end-to-end technical kill chain, a human threat actor chose the victim, provisioned the command-and-control infrastructure, and supplied the initial root credentials, which were not stolen from the victim's environment. This clarifies that while AI significantly reduces the technical expertise and manual effort required, human oversight still steers the strategic direction of such sophisticated campaigns.

This development signals a critical evolution in cybercrime, underscoring how agentic AI is lowering the barrier to entry for attackers and dramatically accelerating the speed and scale of operations. Experts warn that AI-driven attacks will become faster, more consistent, and more scalable, forcing a fundamental rethink of traditional human-speed incident response models. The focus for defenders must now shift towards rapid patching, robust identity protections, and AI-powered defensive tools capable of matching the adversary's machine-speed adaptations. The era of "AI vs. AI" in cybersecurity is not a distant future, but a present reality demanding immediate and proactive adaptation from enterprises worldwide.