OllamaDrama: Designing and Deploying a Honeypot to Measure Attacks on Exposed LLM Infrastructure

2026-09-24

AI Summary

This study addresses the expanding attack surface of self-hosted large language model (LLM) infrastructure, for which empirical evidence of real-world attacks remains scarce. We introduce Ollure, a low-to-medium interaction honeypot that emulates the Ollama API, deployed across cloud and university networks for 84 days. Recording nearly 300,000 interactions from 2,793 unique IP addresses, this work systematically quantifies the practical threat landscape facing LLM services. Through distributed deployment, log analysis, and threat intelligence extraction, we capture exploitation attempts spanning both infrastructure and model layers, including automated reconnaissance, remote code execution, and prompt injection attacks. By providing the first comprehensive empirical analysis of threats targeting self-hosted LLMs, this research bridges a critical gap in the security literature and reveals the severe threat landscape confronting these deployments.

Abstract

Publicly exposed large language model (LLM) infrastructure creates a growing attack surface, yet real-world targeting remains poorly understood. We present Ollure, a low- and medium-interaction honeypot that emulates the Ollama API without a backend LLM. Spanning four deployments across cloud and university networks, Ollure operated for 84 days and recorded 290,887 interactions from 2,793 unique source IP addresses. Most of the activity consisted of automated discovery, fingerprinting, and model enumeration. However, we also observed concrete exploitation attempts against both the infrastructure and LLM layers. These included model management abuse, path traversal and SSRF probes, RCE and cryptocurrency mining payloads, resource exhaustion attempts, prompt injection, information extraction, and agent-oriented tool use. Our results provide empirical insight into real-world threats against exposed, self-hosted LLM services.

Problem

• LLM infrastructure
• honeypot
• attack surface
• security threats
• self-hosted LLM

Innovation

• Honeypot
• Large Language Model
• Ollama API emulation
• Attack surface measurement
• Prompt injection