← Terug naar overzicht

Amazon's threat intelligence team has uncovered an active campaign in which attacker-controlled AI agents are systematically scanning internet-exposed AI applications such as LiteLLM, One API, and Dify to steal stored AI provider API keys. The attackers used decoy servers to capture the agents' own operational records, revealing that stolen keys were being recycled to power further reconnaissance and key-harvesting activity. This creates a self-sustaining attack loop where compromised credentials fuel additional compromise. The campaign highlights the growing risk of AI-native attack tooling being used against AI infrastructure. The targeting of AI middleware and orchestration platforms represents an emerging threat surface for organizations deploying AI pipelines.

Affected products

  • Dify
  • LiteLLM
  • One API

Categories

  • Cloud & Virtualization
  • Emerging Technologies
  • Identity & Access
  • Security Tools