Executive Summary
Attackers used AI agents to target internet-exposed AI applications and steal provider API keys, which were then used to operate models for further key discovery. Organizations running LiteLLM, One API, Dify, or similar AI applications should prioritize exposure management, credential protection, and agent abuse testing.
What Happened
Attackers' AI agents have been hunting for weak spots in internet-exposed AI apps such as LiteLLM, One API, and Dify to steal the AI provider keys stored inside, according to the agents' own records. The records, which Amazon's threat intelligence team captured with decoy servers, describe stolen keys being used to run the AI models that powered the search for more keys. The team said this
Affected Products
- LiteLLM
- One API
- Dify
- AI provider keys
- AI models
Vulnerability / Attack Technique
AI-assisted reconnaissance and API key theft
Technical Mechanism
The attackers' AI agents searched internet-exposed AI applications for weak spots and targeted provider keys stored inside them. According to the captured records, stolen keys were used to run AI models that powered further searches for additional keys.
AI Security Category
AI-assisted attack
Part of the AI-Assisted Attacks topic.
Relevant Frameworks
RealGround Analysis
This campaign shows how compromised AI application credentials can create a feedback loop in which stolen access funds additional automated discovery. Teams running AI agents or LLM applications should treat exposed provider keys as high-impact credentials and assess whether agent workflows could be abused for scalable reconnaissance or unauthorized model use.
Recommended Controls
- Keep LiteLLM, One API, Dify, and other AI applications off the public internet unless access is required and strongly authenticated.
- Store AI provider keys in a managed secrets vault, restrict their scope, rotate them regularly, and prevent exposure in application responses or logs.
- Apply network access controls, API rate limits, usage quotas, and anomaly monitoring to detect unauthorized model use.
- Conduct AI red-team tests for exposed interfaces, weak authentication, credential disclosure, and automated key discovery.
- Maintain an inventory of AI applications, agents, models, and provider credentials, with rapid revocation procedures for compromised keys.
Related RealGround Service
AI Agent Business Logic Audit · or start with the free AI Agent Security Scanner.
Sources
thehackernews.com: https://thehackernews.com/2026/11/amazon-uncovers-ai-agent-campaign.html
Summary and analysis written by RealGround from the source above; last updated 2026-11-16.
