CEFET/RJ (Centro Federal de Educação Tecnológica Celso Suckow da Fonseca): 17-Model Brazilian Portuguese AI Stack, NuClide engagement record

NuClide Research · 2026-05-01

Summary

Brazil’s CEFET/RJ (Federal Center for Technological Education Celso Suckow da Fonseca) has an Ollama instance with 17 models, including custom Brazilian Portuguese fine-tunes and a 39GB DeepSeek-R1:70B local model. No authentication. Heavy emphasis on Portuguese-language AI research/coursework.

Infrastructure

Field	Value
IP	200.9.149.153
Organization	Centro Federal de Educação Tecnológica Celso Suckow da Fonseca (CEFET/RJ)
Country	Brazil
Open ports	11434 (Ollama, public)

Model Inventory (17 models)

Model	Size	Notes
`deepseek-r1:70b`	39GB	Local DeepSeek-R1 70B
`RecognaNLP/chatbode:7b`	14GB	Brazilian Portuguese chatbot (RecognaNLP, USP/UFSCar lab)
`cnmoro/mistral_7b_portuguese:q2_K`	2GB	Portuguese-fine-tuned Mistral
`lukashabtoch/plutotext-r3-emotional:latest`	4GB	Custom: “emotional text” model
`lukashabtoch/moirai-agent:latest`	1GB	Custom: agent model (likely student project)
`mattw/pygmalion:latest`	3GB	Pygmalion (roleplay/chat model)
`mario:latest`	1GB	Custom small model
`gemma:7b`	4GB	,
`mistral:7b`	4GB	,
`llama3.2:3b-instruct-q5_K_M`	2GB	,
`llama3.1:latest`	4GB	,
`llama2:latest`	3GB	,
`llama3-backup:latest`	1GB	,
`llama3.2:latest`	1GB	,
`tinyllama:latest`	,	,
`smollm2:135m`	,	,
`deepseek-r1:latest`	4GB	,

Findings

F1: Unauthenticated Inference Across 17 Models (HIGH)

All 17 models accessible without authentication. The 39GB DeepSeek-R1:70B requires significant GPU resources, free inference at CEFET/RJ’s compute expense.

F2: Custom Brazilian Portuguese AI Models Exposed (HIGH)

Three custom-namespace models indicate either student/faculty research projects or coursework deployments:

lukashabtoch/moirai-agent, likely a student’s “Moirai” agent project
lukashabtoch/plutotext-r3-emotional, emotional text model (R3 = release 3)
mario:latest, unnamed researcher’s model

CVE-2025-63389 injection on these custom models would silently affect any student/researcher relying on their outputs.

The RecognaNLP/chatbode:7b model is the public Brazilian Portuguese ChatBode (from USP/UFSCar’s RecognaNLP group), its system prompt confirms instructional/conversational use:

“Você é assistente de IA chamado ChatBode… projetado para ser prestativo, honesto e inofensivo.”

F3: Pygmalion Roleplay Model Present (MEDIUM)

mattw/pygmalion:latest is a roleplay/character-chat model, unusual deployment on a federal educational institution server.

Remediation

OLLAMA_HOST=127.0.0.1:11434
systemctl restart ollama

Disclosure

Discovered: 2026-05-01
Status: Pending outreach to CEFET/RJ NTI (Núcleo de Tecnologia da Informação)