ENCFORGE Ransomware Exploits Unpatched Langflow RCE to Encrypt AI Assets

The JADEPUFFER threat actor is actively exploiting CVE-2025-3248, a critical remote code execution vulnerability in older versions of Langflow, to deploy the ENCFORGE ransomware against AI infrastructure. This attack is particularly damaging because it specifically targets high-value AI assets — model weights and training datasets — which are often irreplaceable and represent enormous investment in compute and data. The root cause is straightforward: organizations running unpatched, internet-facing Langflow instances gave attackers a trivial entry point requiring no credentials. This incident highlights that AI development platforms carry the same exposure risks as any other internet-facing service and must be included in standard patch and vulnerability management programs.

Tactical Insight

Immediate actions

Patch all Langflow instances to the latest version to remediate CVE-2025-3248 immediately.
Restrict public internet access to Langflow endpoints by placing them behind a VPN or zero-trust access gateway.
Audit and disable any unnecessary or unauthenticated API endpoints exposed by AI development platforms.

Long-term improvements

Integrate AI development tools (Langflow, MLflow, etc.) into your formal vulnerability management and asset inventory program.
Implement immutable, air-gapped backups of AI model weights and training datasets to enable recovery without paying ransom.
Apply network segmentation to isolate AI infrastructure from production systems and other sensitive environments.

Detection measures

Deploy file integrity monitoring on directories containing AI model files to alert on unexpected mass encryption or renaming activity (e.g., `.locked` extension).
Monitor Langflow and AI platform logs for anomalous API calls, unusual process spawning, or outbound connections indicative of ransomware staging.
Establish behavioral detection rules in your EDR/SIEM for Go-based binaries executing bulk file encryption patterns.

Framework References

CIS Control 7: Continuous Vulnerability ManagementCIS Control 12: Network Infrastructure ManagementCIS Control 11: Data RecoveryCIS Control 13: Network Monitoring and DefenseNIST CSF ID.AM-2: Software platforms and applications within the organization are inventoriedNIST CSF PR.IP-12: A vulnerability management plan is developed and implementedNIST SP 800-53 SI-2: Flaw RemediationNIST SP 800-53 SC-7: Boundary ProtectionNIST SP 800-53 CP-9: System BackupNIST SP 800-53 AU-6: Audit Record Review, Analysis, and ReportingMITRE ATT&CK T1190: Exploit Public-Facing ApplicationMITRE ATT&CK T1486: Data Encrypted for ImpactITIL: Change and Release Management — Emergency Patching Procedures

Source

ThreatNoir coverageNew ENCFORGE Ransomware Targets AI Model Files in Langflow RCE Attack Read originalNew ENCFORGE Ransomware Targets AI Model Files in Langflow RCE Attack

Heard on the podcastThreatNoir Afternoon Brief — July 21

PublishedJul 21, 2026