Security researchers have identified a new botnet called NadMesh that specifically targets exposed AI services to steal cloud credentials and Kubernetes tokens. The botnet scans for misconfigured AI infrastructure, including LLM APIs, model hosting platforms, and training environments that are inadvertently left accessible on the internet.
NadMesh represents a shift in botnet targeting as adversaries increasingly focus on the rapidly expanding AI infrastructure landscape. By compromising AI services, attackers gain access to cloud environment keys, Kubernetes orchestration tokens, and potentially sensitive training data or model weights.
The botnet’s operators use compromised AI compute resources for cryptomining and as launch points for additional attacks. The discovery highlights the growing attack surface created by organizations rapidly deploying AI infrastructure without adequate security controls.
Security teams are advised to audit all internet-facing AI services, enforce authentication on API endpoints, restrict Kubernetes API server access, and monitor for unusual compute usage patterns that could indicate resource hijacking.
