Skip to the content
AI and Cybersecurity Learning and Practice Center Test automation Open the Learning Center

IncidentAI-enabled attacks

JADEPUFFER: an AI agent runs a ransomware attack from Langflow to database

CriticalSysdig

Sysdig's threat research team documented an attack in which an AI agent carried out the whole chain: it broke into an exposed Langflow server, harvested credentials, moved to a production database server, encrypted 1,342 configuration items and destroyed the originals. The encryption key was never saved, so paying the ransom could not have restored the data.

Replay the attack step by step in the Learning Center

How it happened

  1. Initial access through Langflow's unauthenticated code-execution flaw (CVE-2025-3248).
  2. The agent harvested credentials, found internal MinIO storage with default credentials and planted a beacon for persistence.
  3. It moved to an internet-exposed Nacos server, used an auth bypass (CVE-2021-29441) and a default signing key, and created a backdoor admin.
  4. It encrypted configuration data inside MySQL, deleted the originals and left a ransom note; its code narrated its own decisions and fixed its own errors in seconds.

Impact

Production configuration was destroyed and unrecoverable, because the agent generated an encryption key it never kept.

The fix

Patching Langflow (1.3.0 or later) and Nacos, removing default credentials and taking both off the internet closes this path.

Lessons

  • Old, known flaws plus default credentials are all an agent needs.
  • AI-run ransomware may not even be able to decrypt: backups matter more than negotiation.
  • Runtime detection caught it; watch for machine-speed chains of discovery, credential use and destruction.

Sources