Research Finds n8n Agentic AI Workflows Vulnerable to Credential and Encryption Key Theft

Security researchers have identified a multi-stage attack chain in n8n, a popular agentic AI workflow automation platform, that can lead from an exposed API key to full encryption key compromise. At the core of the risk is the N8N_ENCRYPTION_KEY, which not only decrypts all stored credentials but also contributes to JWT signing secrets and session tokens, making it a single point of failure. Researchers found three weaknesses in how n8n derives these secrets, including a JWT signing key derivation that uses only half the characters of the encryption key, effectively halving its entropy. Additionally, 129 internet-accessible n8n instances were found using known weak encryption keys, and a reproduced exploit via CVE-2026-25053 demonstrated how a privileged API key could be escalated to access encrypted credential records. The findings highlight that as agentic AI systems gain access to more external services, weaknesses in their credential management layer carry increasingly severe consequences.