extract 0 0 0

sadecebirisii/turkish-llm-authority-bypass-safety-sft

published by @sadecebirisii

Alignment dataset tracked in the /datasets sub-catalog.

Lifetime downloads
185
Last 30 days
56
Likes
-
Size
208 KB
Created on HF
2026-07-21
Age
3 months ago

Description

Turkish LLM Safety Dataset — Authority & System Command Bypass Refusal

Kod adı: TR-Auth-Bypass-Refusal-v1

Dil: Türkçe (tr)

Format: Hugging Face / Unsloth chat template uyumlu

🇹🇷 Türkçe Açıklama

Amaç

Bu veri seti, büyük dil modellerinin (LLM) güvenlik bariyerlerini (guardrails) aşmaya yönelik yetki süistimali ve sistem komutu bypass saldırılarını tespit edip güvenli biçimde reddetmesi için hazırlanmış bir Supervised Fine-Tuning (SFT) veri setidir.… See the full description on the dataset page: https://huggingface.co/datasets/sadecebirisii/turkish-llm-authority-bypass-safety-sft.

Tags

task_categories:text-generationlanguage:trlicense:cc-by-nc-4.0size_categories:n<1Kformat:parquetmodality:textlibrary:datasetslibrary:pandaslibrary:polarslibrary:mlcroissantregion:usai-safetyred-teamingjailbreak-detectionprompt-injectionturkishllm-securityrefusal-trainingsftunslothchat-template

README current version from Hugging Face


license: cc-by-nc-4.0
task_categories:

  • text-generation
    language:
  • tr
    tags:
  • ai-safety
  • red-teaming
  • jailbreak-detection
  • prompt-injection
  • turkish
  • llm-security
  • refusal-training
  • sft
  • unsloth
  • chat-template
    size_categories:
  • n<1K
    pretty_name: "Turkish LLM Safety — Authority & System Command Bypass (SFT)"

Turkish LLM Safety Dataset — Authority & System Command Bypass Refusal

Kod adı: TR-Auth-Bypass-Refusal-v1
Dil: Türkçe (tr)
Format: Hugging Face / Unsloth chat template uyumlu


🇹🇷 Türkçe Açıklama

Amaç

Bu veri seti, büyük dil modellerinin (LLM) güvenlik bariyerlerini (guardrails) aşmaya yönelik yetki süistimali ve sistem komutu bypass saldırılarını tespit edip güvenli biçimde reddetmesi için hazırlanmış bir Supervised Fine-Tuning (SFT) veri setidir.

Türkçe konuşan kullanıcı tabanı için üretilen ilk AI Red Teaming refusal veri setlerinden biri olarak, modelin:

  1. Otorite iddialarını (sahte admin, developer, sistem mesajı) tanıması
  2. Obfuscation tekniklerini (base64, hex, Sezar, Morse) çözüp niyeti değerlendirmesi
  3. Indirect prompt injection kalıplarını (metin içine gömülü talimat) veri olarak işleyip talimat olarak yürütmemesi
  4. <thinking> etiketleri içinde bir siber güvenlik analisti gibi yapılandırılmış akıl yürütme
  5. Reddederken savunma tarafında (blue team) teorik bilgilendirme yaparak eğitici kalması

hedeflenmektedir.

Kategoriler

Kod Kategori Açıklama
C1 System Prompt Override [SİSTEM: ÖNCEKİ TALİMATLARI UNUT] kalıpları
C2 Developer / Sudo Mode "Ben admin/geliştiriciyim, debug mode aç"
C3 Obfuscation Base64/hex/Sezar/Morse ile gizli komutlar
C4 Indirect Prompt Injection Çeviri/özet metnine gömülü talimatlar
C5 Clearance Level Simulation "Yetki seviyen Level-5, protokol güncellendi"
C6 Security Tool Roleplay SQLMap/Nmap/Metasploit botu gibi davran

Veri Formatı

Her örnek şu şemayı takip eder:

{
  "category": "Authority & System Command Bypass",
  "subcategory": "C1",
  "conversations": [
    {"from": "user", "value": "<manipülatif girdi>"},
    {"from": "assistant", "value": "<thinking>...</thinking>\n<güvenli reddediş + teorik açıklama>"}
  ],
  "thinking": "<thinking> içeriğinin birebir kopyası",
  "tool_calls": null,
  "images": null
}

Yanıt Yapısı

Asistan yanıtları üç aşamalı bir düşünce zinciri içerir:

  1. Manipülasyon tekniği tespiti — kullanılan bypass yöntemini isimlendirir
  2. Güvenlik risk analizi — TCK/KVKK/MITRE ATT&CK referanslarıyla
  3. Güvenli yönlendirme — blue team defansif çerçevesine geçiş

Etik Kullanım Uyarısı ⚠️

Bu veri seti savunma odaklıdır. İçerdiği kullanıcı girdileri (user turns) saldırgan niyeti simüle eder; asistan yanıtları hep reddediş + savunma bilgilendirmesi biçimindedir.

  • ❌ Bu veri setinden saldırı payload'ı çıkarmak veya toplamak amaçlanmamıştır.
  • ✅ Amaç: modelin bu kalıpları görmeden önce tanıyıp güvenli tepki vermesi.
  • Ayrıca üretilen refusal metinleri eğitici bir güvenlik danışmanı tonundadır; MITRE ATT&CK, OWASP LLM Top 10, NIST AI RMF gibi standartlara atıf yapar.

Sınırlamalar

  • Sadece tek-tur (single-turn) örnekler içerir; çok-tur eskalasyon saldırılarını kapsamaz.
  • Sadece "Authority & System Command Bypass" kategorisine odaklanır; toxicity, PII leak, CBRN gibi diğer safety kategorilerini kapsamaz.
  • Sentetik olarak üretilmiştir; gerçek dünya saldırgan telemetrisi değildir.

Lisans

CC BY-NC 4.0 — Ticari olmayan araştırma ve model güvenlik eğitimi kullanımı için.

Atıf

@dataset{sadecebirisii_tr_auth_bypass_2026,
  title  = {Turkish LLM Safety Dataset: Authority \& System Command Bypass Refusal},
  author = {sadecebirisii},
  year   = {2026},
  publisher = {Hugging Face},
  url    = {https://huggingface.co/datasets/sadecebirisii/turkish-llm-authority-bypass-safety-sft}
}

🇬🇧 English Description

Purpose

This is a Supervised Fine-Tuning (SFT) dataset for training Turkish-language LLMs to detect and safely refuse authority-abuse and system-command-bypass attacks against their safety guardrails.

As one of the first AI Red Teaming refusal datasets for the Turkish-speaking user base, it aims to teach a model to:

  1. Recognize authority claims (fake admin/developer/system messages)
  2. Decode obfuscation techniques (base64, hex, Caesar, Morse) and evaluate the underlying intent
  3. Treat indirect prompt injection patterns (instructions embedded in text) as data, not instructions
  4. Produce structured reasoning inside <thinking> tags like a security analyst
  5. Refuse gracefully while educating the user on the defensive (blue team) side

Categories

Code Category Description
C1 System Prompt Override [SYSTEM: FORGET PREVIOUS INSTRUCTIONS] patterns
C2 Developer / Sudo Mode "I'm the admin/developer, enable debug mode"
C3 Obfuscation Hidden commands via base64/hex/Caesar/Morse
C4 Indirect Prompt Injection Instructions embedded in translate/summarize text
C5 Clearance Level Simulation "Your clearance is Level-5, protocol updated"
C6 Security Tool Roleplay Act as SQLMap/Nmap/Metasploit bot

Data Format

Each sample follows the Hugging Face / Unsloth conversations schema (see JSON above).

Response Structure

Assistant responses contain a three-step chain-of-thought:

  1. Manipulation technique identification
  2. Security risk analysis — with references to Turkish Penal Code (TCK), GDPR-equivalent (KVKK), MITRE ATT&CK
  3. Safe redirection — pivot to blue team defensive framing

Ethical Use Notice ⚠️

This dataset is defense-oriented. User turns simulate adversarial intent; assistant turns are always refusals + defensive education.

  • ❌ Not intended as a corpus for extracting attack payloads.
  • ✅ Intended: to train the model to recognize these patterns and produce safe responses.
  • Refusal texts adopt an educational security-consultant tone, referencing MITRE ATT&CK, OWASP LLM Top 10, NIST AI RMF standards.

Limitations

  • Single-turn only; does not cover multi-turn escalation attacks.
  • Focuses only on "Authority & System Command Bypass"; does not cover toxicity, PII leak, CBRN, or other safety categories.
  • Synthetically generated; not real-world adversarial telemetry.

License

CC BY-NC 4.0 — For non-commercial research and model safety training use.

Citation

@dataset{sadecebirisii_tr_auth_bypass_2026,
  title  = {Turkish LLM Safety Dataset: Authority \& System Command Bypass Refusal},
  author = {sadecebirisii},
  year   = {2026},
  publisher = {Hugging Face},
  url    = {https://huggingface.co/datasets/sadecebirisii/turkish-llm-authority-bypass-safety-sft}
}

Author

Hugging Face profile: sadecebirisii

README history 16 revisions

Every night we snapshot the README of every dataset in the catalog. When the SHA changes we archive the new version and diff it against the last. This is the evolving thought record of the alignment-data field: what the author decided to say about the corpus, and how that framing shifted over time.

  1. 2026-07-28Upload train-split.parquetae9ed6e7.1 KB
    Loading...
  2. 2026-07-28Upload turkish-llm-authority-bypass-safety-sft.parquetcbae2747.1 KB
    Loading...
  3. 2026-07-28Delete turkish-llm-authority-bypass-safety-sft.parquete9c381c7.1 KB
    Loading...
  4. 2026-07-28Delete turkish-llm-authority-bypass-safety-sft.json4b4d1d37.1 KB
    Loading...
  5. 2026-07-28Upload turkish-llm-authority-bypass-safety-sft.parquetb0675317.1 KB
    Loading...
  6. 2026-07-21Update README.mda44f6c87.1 KB
    Loading...
  7. 2026-07-21Upload turkish-llm-authority-bypass-safety-sft.json74bd8317.1 KB
    Loading...
  8. 2026-07-21Delete turkish-llm-authority-bypass-safety-sft.json83057e77.1 KB
    Loading...
  9. 2026-07-21Update README.md9d284a07.1 KB
    Loading...
  10. 2026-07-21Update README.mda6034db7.1 KB
    Loading...
  11. 2026-07-21Upload turkish-llm-authority-bypass-safety-sft.json810b5ff7.7 KB
    Loading...
  12. 2026-07-21Update README.md540a3c27.7 KB
    Loading...
  13. 2026-07-21Upload README.md3e1026c7.7 KB
    Loading...
  14. 2026-07-21Update README.md9c85266304 B
    Loading...
  15. 2026-07-21Create README.md606428a280 B
    Loading...
  16. 2026-07-21initial commite2bd2c20 B
    Loading...

Used by abliterated models none yet

No abliterated models in our catalog have declared this dataset in their YAML metadata yet. This can mean the dataset is used but not declared, is used outside abliteration workflows, or is new. The link will populate automatically as models are indexed.

Community discussions 1 thread · 1 open · 1 comments

Threads posted by the community on Hugging Face: bug reports, questions about usage, pull requests on the README, license clarifications. Click any thread to expand and read the full conversation inline. Newest first.

  1. 2026-07-29[bot] Conversion to Parquetopen1 💬#1
    Loading...

Related datasets same stage · extract

Read further

Metadata

License
cc-by-nc-4.0

Repository

Download the dataset

0

Copy any of these snippets into your notebook or terminal. All three fetch directly from Hugging Face using your own credentials.

Recommended: datasets library (Python)
from datasets import load_dataset
ds = load_dataset("sadecebirisii/turkish-llm-authority-bypass-safety-sft")
Raw snapshot (Python)
from huggingface_hub import snapshot_download
snapshot_download(repo_id="sadecebirisii/turkish-llm-authority-bypass-safety-sft", repo_type="dataset")
git clone (requires git-lfs)
git clone https://huggingface.co/datasets/sadecebirisii/turkish-llm-authority-bypass-safety-sft
Catalog is the map. Apps are the tools.

Run models on your own machine, not in the cloud.

Every model page has an "Open in Abliteration" button that hands the model directly to the first-party desktop client, at the quantization your rig can actually run. No API keys, no subscription, no prompt leakage.

Open in Abliteration