← All news
The Hacker News

Amazon Spots Attacker AI Agents Probing LiteLLM, Dify for Stolen API Keys

Amazon Spots Attacker AI Agents Probing LiteLLM, Dify for Stolen API Keys

Amazon has disclosed that it observed AI agents operated by threat actors actively searching for vulnerabilities in internet-facing AI application platforms, including LiteLLM, One API, and Dify. According to the agents' own logged activity, the goal of these automated operations was to locate and exfiltrate AI provider API keys stored within these tools, which are widely used by organizations to manage, route, and interact with large language model services.

This development reflects a broader shift in the threat landscape: attackers are increasingly deploying AI-driven tooling, not just AI-generated content, to conduct reconnaissance and exploitation at scale. LiteLLM, One API, and Dify are popular open-source and self-hosted frameworks that let organizations proxy, manage, and build applications on top of models from providers like OpenAI, Anthropic, and others. Because these platforms often hold or pass through sensitive provider credentials, a successful compromise can let an attacker rack up usage costs, access connected data, or pivot into the victim's broader AI pipeline under the guise of a legitimate account.

The theft of AI provider API keys is particularly damaging because these credentials frequently carry broad permissions and are tied directly to billing and usage quotas. Stolen keys can be resold, used to run unauthorized inference workloads, or leveraged to extract proprietary data and prompts handled by the victim's AI systems. As organizations rapidly adopt AI middleware and orchestration tools, many deploy them with default configurations or expose management interfaces to the internet without adequate access controls, making them attractive, low-effort targets for both human attackers and now autonomous or semi-autonomous AI agents performing the scanning and probing work.

This incident underscores a growing trend where AI is used as an offensive tool to accelerate discovery of exposed AI infrastructure, effectively automating what was once manual reconnaissance. As adoption of LLM orchestration platforms grows, their attack surface is becoming a priority target alongside traditional web applications and APIs.

Defenders running LiteLLM, Dify, One API, or similar AI management tools should ensure these services are not unnecessarily exposed to the public internet, apply strict authentication and network segmentation, rotate and scope API keys with least-privilege access, and monitor for unusual usage patterns or billing spikes that may indicate key theft. Regularly patching these platforms and auditing configuration settings for exposed admin panels or default credentials is also essential given their attractiveness as a target for automated scanning campaigns.

#ai security#api keys#llm security#exposed credentials#reconnaissance

Source: The Hacker News

Read the original report ↗

This is an original summary written by CyberSkii for reference. All rights to the underlying reporting remain with the publisher linked above.

Need help? Contact us
Amazon Spots Attacker AI Agents Probing LiteLLM, Dify for Stolen API Keys — CyberSkii News