Researchers at Sysdig have connected a second attack on a Langflow server to JADEPUFFER, an AI-agent-driven threat actor documented earlier this month. The group deployed ENCFORGE, a newly discovered ransomware written in Go, targeting AI infrastructure rather than traditional business data.

ENCFORGE encrypts model weights, vector indexes, training datasets, and other AI-specific files across infected systems. This represents a shift in ransomware strategy. Attackers now recognize the value of machine learning assets, which often lack backup protection and carry substantial recovery costs.

The attacks exploited Langflow, an open-source platform for building AI applications. Entry vectors remain unconfirmed in available reporting, though Langflow has disclosed remote code execution vulnerabilities in recent versions. Organizations running unpatched instances face direct exposure.

JADEPUFFER differs from traditional ransomware operators. Rather than deploying malware manually, the group uses AI agents to coordinate attack sequences autonomously. This automation reduces reliance on human operators and accelerates lateral movement across networks.

The targeting of AI model files creates distinct organizational risks. Unlike structured databases, machine learning models often exist in distributed locations without centralized protection. Recovery requires either paying the ransom or reconstructing models from scratch, a process consuming weeks or months depending on training complexity.

Organizations running Langflow deployments should prioritize patching immediately. Network monitoring for ENCFORGE activity centers on detecting Go-compiled binaries with file encryption behavior. The ransomware's focus on specific file extensions makes it identifiable through filesystem monitoring tools.

Backups remain the primary defense. AI infrastructure teams must implement offline copies of model weights and training datasets, stored separately from production systems. Standard ransomware resilience practices apply here, but the specialized nature of AI assets means generic backup strategies may miss critical files.

The emergence of ENCFORGE demonstrates that ransomware