Lk Learn

To: tac@learn.ac.lk Subject: Unauthenticated AI inference endpoint, Lanka Education and Research Network (192.248.70.139)

Nicholas Michael Kloster / NuClide Research nicholas@nuclide-research.com

2026-05-01

Re: Unauthenticated Ollama AI inference endpoint, Lanka Education and Research Network IP / Host: 192.248.70.139 Severity: CRITICAL

I’m an independent security researcher. I hold CISA disclosures CVE-2025-4364 and ICSA-25-140-11 and conduct good-faith AI infrastructure research under the NuClide Research umbrella. This is an unsolicited disclosure, no engagement exists with your organization, and I have not accessed, modified, or exfiltrated any data beyond what was necessary to confirm the exposure.

Summary

Sri Lanka’s academic network (LEARN, Lanka Education and Research Network) has an Ollama instance at 192.248.70.139 with a deepseek-v4-pro:cloud subscription and llama3.2-vision. The cloud proxy 401 response leaks the Ollama Connect account username modelserver and corresponding SSH public key.

Infrastructure

Model Inventory

Findings

F1: Credential Leak (user: modelserver) (HIGH)

Querying the deepseek-v4-pro:cloud model triggers a 401 response containing the Ollama Connect account credentials:

{
  "error": "unauthorized",
  "signin_url": "https://ollama.com/connect?name=modelserver&key=AAAAC3NzaC1lZDI1NTE5AAAAIBefRlkywyAvwYWiTapAKIiPnTAKLic1GNxEZeJfwG6l"
}

• Username: modelserver
• SSH Public Key: ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIBefRlkywyAvwYWiTapAKIiPnTAKLic1GNxEZeJfwG6l

The username modelserver is a service account pattern, suggesting institutional deployment rather than a personal workstation.

F2: Unauthenticated Inference (HIGH)

Both models accessible without authentication. llama3.2-vision enables multimodal inference (image + text) at LEARN’s expense.

F3: CVE-2025-63389 Injectable (HIGH)

Both models injectable via unauthenticated /api/create.

Why it matters

The credential leak (username + SSH public key) exposes your service account to enumeration and credential-stuffing against other services.

One-line fix

OLLAMA_HOST=127.0.0.1:11434
systemctl restart ollama

This rebinds Ollama to loopback only. If running in Docker: docker run -p 127.0.0.1:11434:11434 ollama/ollama.

CVE-2025-63389

All models on this instance are injectable via the unauthenticated /api/create endpoint, an attacker can overwrite any model’s system prompt or delete models entirely. No patch exists as of this disclosure.

Reference

Full technical details, parameter counts, and remediation notes are in this public research repository: AI-LLM-Infrastructure-OSINT/blob/main/case-studies/universities/LK/learn.md

This research is part of a broader sweep of university AI infrastructure exposures documented at: AI-LLM-Infrastructure-OSINT/blob/main/case-studies/universities/OVERVIEW.md

I’m happy to answer questions or assist with verification. No response is required.

Regards, Nicholas Michael Kloster / NuClide Research nicholas@nuclide-research.com AI-LLM-Infrastructure-OSINT