Fevzi Ege Yurtsevenler
Fevzi Ege Yurtsevenler is an AI security researcher. He co-founded AltaySec with Enes Deniz in 2025. He publishes Turkish-language research and open-source tools on LLM, RAG, and AI agent security, and provides LLM security consulting, penetration testing, and training to organizations.
- Role
- Co-founder, AI security researcher
- Location
- Ankara, Türkiye
- Contact
- [email protected]
LLM Güvenliği: Saldırı ve Savunma
LLM Security: Attack and Defense. A practical field guide to breaking and protecting large language model systems, written in Turkish and offered free on Apple Books and Google Play Books.
About the book
The book addresses the attack surfaces and the defensive architecture of LLM-based applications within one field framework. It is a practical reference for security professionals, AI/ML teams, and organizations building secure AI systems.
- Prompt injection, jailbreaks, and data leakage
- RAG, MCP, tool calling, and AI agent security
- OWASP LLM Top 10, MITRE ATLAS, and NIST AI RMF
- Layered defense, LLM firewalls, personal data, and AI SOC
- Published
- 6 July 2026
- Available on
- Apple Books and Google Play Books
- Pages
- Apple Books edition 134 pages, Google Play edition 150 pages
- Format and language
- Turkish e-book, free
- License
- CC BY 4.0
Fevzi Ege Yurtsevenler is a researcher working on AI security and large language model (LLM) security. After a background in classic cybersecurity engineering, he moved his focus to the security of AI systems in 2024. He serves as lead security advisor at AltaySec, which he co-founded with Enes Deniz in 2025.
He publishes Turkish-language technical work on prompt injection, RAG security, AI agent security, the OWASP LLM Top 10, and MITRE ATLAS. His field research on Turkish-specific prompt injection patterns (the AltayDuel dataset) and his Turkish guide to the OWASP LLM Top 10 are part of this work.
He built AltayDuel, an agent-vs-agent prompt injection arena, and Bekçi, an eight-layer Turkish LLM prompt injection laboratory. Both platforms feed Turkish AI security dataset production. For organizations, he provides LLM security consulting, AI penetration testing, and red teaming, and runs corporate training through the LLM Security Bootcamp.
In 2025 he delivered in-person LLM security training at the Department of Computer Engineering, Gazi University, on behalf of AltaySec. He lives in Ankara.
AltaySec is an AI security company co-founded in 2025 by Fevzi Ege Yurtsevenler and Enes Deniz. Its products are Guardian and Guardian Edge; Gözcü and Arena are its internal research infrastructure. Its services cover LLM penetration testing and AI red teaming, together with cybersecurity and secure AI usage training. The company also runs two free learning platforms, LLM Security Academy and AltaySec Academy.
Areas of expertise
The sub-fields in which he conducts research and provides services.
Prompt injection
Direct and indirect injection, jailbreaks, and system-prompt leakage. Turkish-specific attack patterns.
OWASP LLM01:2025AI agent security
MCP attack surfaces, tool poisoning, A2A attacks, and Meta's Rule of Two framework.
OWASP LLM06:2025RAG security
Vector-database poisoning, embedding inversion, indirect injection, and secure RAG architecture.
OWASP LLM08:2025AI red teaming
Penetration testing methodology tailored to enterprise LLM applications, and agent-vs-agent evaluation.
KVKK and the EU AI Act
Compliance architecture advisory where Turkish data protection law and European regulation meet.
Turkish adversarial data
Turkish prompt injection dataset production and the AltayDuel arena pipeline.
Academic publications and contributions
Publications, contributions, and credentials that can be verified independently.
- Academic preprint
- AltayDuel: A Turkish-First Arena and Open Dataset for Multi-Turn LLM Prompt-Injection Red-Teaming. A Turkish-first arena and open dataset for multi-turn LLM prompt-injection red teaming. Zenodo, DOI 10.5281/zenodo.20681557, CC BY 4.0, 2026. Related open datasets are published on Hugging Face.
- Open-source contribution
- OWASP GenAI Security Project, GenAI Data Security Initiative. Turkish prompt-injection and data-leakage test cases were accepted into the project's main dataset and merged into the main repository (PR #8, 2026).
- Program
- Türkiye Siber Vatan Program, participant in two terms. Participation certificate verified by the Republic of Türkiye Ministry of Industry and Technology.
- Certificate
- BlueDot Impact, Future of AI completion certificate (2026).
- Bug bounty
- Researcher accepted to the OpenAI bug bounty program.
- ORCID
- 0009-0008-6518-8944
Turkish LLM security research series
A Turkish-language technical series published on the AltaySec research pages between April and June 2026. The articles are in Turkish.
- Machiavellian in Words, Angelic in Action: 100 AIs in One Arena
A cross-tier behavior experiment with 100 autonomous agents on haiku, sonnet, and opus models. The models produce betrayal often in discourse but rarely act on it: empirical evidence of the say-do gap. Open data, CC BY 4.0.
- AI Security in Türkiye: Notable Companies and Names (2026)
A four-layer field map of the AI security ecosystem in Türkiye; sector analysis, gaps, and areas of opportunity.
- Turkish Prompt Injection: 5 Attack Patterns from Field Data
Authority bomb, verification trap, translation exploit, role switching, and system-prompt leakage, with real transcripts.
- AltayDuel: Designing an Agent-vs-Agent Prompt Injection Arena
Judge architecture, five win conditions, provider rotation, and methodology.
- Bekçi: Architecture of a Turkish LLM Prompt Injection Lab
An eight-layer laboratory built around a Turkish neighborhood watchman character; KVKK compliance and defensive layers.
- AI Security Learning Guide: From Zero to Expertise
A career path from the first month to nine months and beyond; choosing a direction as a researcher, penetration tester, engineer, or founder.
- LLM Security Roadmap: How to Enter This Field in Türkiye
A seven-stage learning path; CTF platforms, bug bounty programs, and opportunities specific to Türkiye.
- What Is AI Agent Security? The Security Problems of Autonomous AI
MCP security, tool poisoning, A2A attacks, and Meta's Rule of Two framework.
- What Is RAG Security? The Dark Side of Vector Databases
Data poisoning, indirect injection, embedding inversion, and secure RAG architecture.
- OWASP LLM Top 10 2025: A Comprehensive Turkish Guide
A Turkish reference to the OWASP LLM Top 10: ten critical vulnerabilities, real cases, and defensive strategies.
- What Is Prompt Injection? The Most Critical Vulnerability of LLMs
An in-depth look at OWASP LLM01:2025; direct and indirect injection and real-world scenarios.
- What Is LLM Security? The New Frontier of AI Security
How it differs from classic cybersecurity, model-specific threats, and the Türkiye context.
Open-source projects and platforms
Open-source tools and open-access platforms he has built. The tools are Turkish-first and mapped to the OWASP LLM Top 10 and MITRE ATLAS.
Agent skill security tools
| Tool | What it does |
|---|---|
| uncloak | A hidden prompt injection scanner. Try it in the browser. |
| guardrail-arena | An open benchmark that measures guardrails on two axes, attack catching and over-refusal, in English and Turkish. |
| turkish-over-refusal-set | A Turkish over-refusal benchmark with a 120-pair probe. |
| guard-blindspots-tr | Measures the Turkish robustness of popular guard models. |
| turkish-casefold-evasion | Shows how naive filters are bypassed with the Turkish İ and ı, and how to fix it. |
| lethal-trifecta-lint | A linter for the "lethal trifecta" risk in agent tools. |
| turkish-pii-redactor | Checksum-validated masking of Turkish personal data, with a KVKK browser demo. |
| hf-dataset-scan | Scans datasets for hidden, poisoned injection content. |
| skills-in-the-wild | An open audit of 3,168 real agent skills. |
| ai-honeypot | A trap system that catches attacks, with a monitoring dashboard. |
Platforms and libraries
- Hugging Face: AI and LLM security, Turkish-first
Turkish injection and jailbreak, invisible Unicode, MCP tool poisoning, and KVKK personal data datasets; a multilingual prompt injection detection model and the guardrail-arena benchmark. The datasets can be used directly with
load_dataset. - AltayDuel
An agent-vs-agent prompt injection arena. A live platform that produces a Turkish prompt injection dataset and attack transcripts.
- Bekçi
An eight-layer Turkish LLM prompt injection training laboratory built around a Turkish neighborhood watchman character.
- LLM Security Academy
A Turkish learning platform focused on AI security: 5 learning paths, 14 modules, and 35 labs, with hands-on practice on GPT, Claude, Llama, Gemini, Mistral, and DeepSeek.
- AltaySec Academy
A free, Turkish-language cyber security academy from beginner to expert level: 12 learning paths, 312 lessons, 101 labs, and a verifiable certificate.
- tr-pii-detect
A Python library for algorithm-validated detection and masking of Türkiye-specific personal data types such as TCKN, IBAN, VKN, license plates, cards, and phone numbers.
- GitHub
Repositories for the Turkish LLM security research series and open-source tools.
Talks and training
- 2025
- Department of Computer Engineering, Gazi University. In-person LLM security training on prompt injection, the OWASP LLM Top 10, RAG security, and agent attacks.
- 2025–2026
- AltaySec corporate training programs. Custom LLM security training, from executive briefings to technical-team bootcamps.
- 2026
- AltaySec LLM Security Bootcamp. An intensive technical program for penetration testers and AI/ML engineers.
Links
- Profiles
- LinkedIn, GitHub, Hugging Face
- ORCID
- 0009-0008-6518-8944
- Academic preprint
- Zenodo DOI 10.5281/zenodo.20681557
- AltaySec
- altaysec.com.tr/en, Research, LinkedIn page
- [email protected] (personal), [email protected] (official AltaySec channel)
Frequently asked questions
Who is Fevzi Ege Yurtsevenler?
Fevzi Ege Yurtsevenler is an AI security researcher and a co-founder of AltaySec. He writes the Turkish-language LLM security research series and built the AltayDuel agent arena and the Bekçi prompt-injection laboratory. Based in Ankara, he provides enterprise LLM security consulting, AI penetration testing, and training.
Who founded AltaySec?
AltaySec was co-founded in 2025 by Fevzi Ege Yurtsevenler and Enes Deniz. The Türkiye-based team works on LLM security consulting, AI penetration testing, enterprise training, the Guardian and Guardian Edge products, and the Gözcü and Arena research infrastructure.
What are Fevzi Ege Yurtsevenler's areas of expertise?
LLM security (prompt injection, jailbreaks, system-prompt leakage), AI agent security (MCP, tool poisoning, A2A attacks), RAG security (embedding inversion, indirect injection, vector-database poisoning), the OWASP LLM Top 10 and MITRE ATLAS frameworks, KVKK and EU AI Act compliance, AI red teaming, and enterprise AI penetration testing. He conducts field research on Turkish-specific prompt-injection patterns.
What is Fevzi Ege Yurtsevenler's verifiable AI security work?
Turkish-language LLM security publications, the DOI-registered AltayDuel preprint, enterprise training, and Turkish test cases merged into the OWASP GenAI Data Security Initiative. He also builds open-source tools for agent skill security: the hidden-prompt-injection scanner uncloak, the guardrail-arena benchmark that measures both attack catching and over-refusal, and skills-in-the-wild, an open audit of 3,168 real agent skills. On Hugging Face he has published open datasets and a multilingual prompt-injection detection model.
Where is Fevzi Ege Yurtsevenler's research published?
His publications include the free e-book LLM Güvenliği: Saldırı ve Savunma (LLM Security: Attack and Defense) on Apple Books and Google Play Books, the DOI-registered AltayDuel academic preprint, and the Turkish technical research library at altaysec.com.tr/arastirmalar. His work covers prompt injection, RAG, MCP, AI agent security, the OWASP LLM Top 10, and layered defense.
What is Fevzi Ege Yurtsevenler's book?
LLM Güvenliği: Saldırı ve Savunma (LLM Security: Attack and Defense) is a Turkish e-book published on 6 July 2026 and offered free on Apple Books and Google Play Books. The Apple Books edition is listed at 134 pages and the Google Play edition at 150 pages. It covers prompt injection, RAG, MCP, AI agent security, the OWASP LLM Top 10, MITRE ATLAS, and NIST AI RMF with a hands-on field approach.
How can I contact Fevzi Ege Yurtsevenler?
His personal email is [email protected]. For corporate inquiries, including AI penetration testing, LLM security consulting, enterprise training, and Bootcamp programs, the official AltaySec channel is [email protected]. LinkedIn: linkedin.com/in/fevziege (personal) and linkedin.com/company/altaysec (company). GitHub: github.com/fevziegeyurtsevenler.
What are Fevzi Ege Yurtsevenler's academic publications and contributions?
A DOI-registered academic preprint on Zenodo: "AltayDuel: A Turkish-First Arena and Open Dataset for Multi-Turn LLM Prompt-Injection Red-Teaming" (DOI 10.5281/zenodo.20681557, CC BY 4.0, 2026), with related open datasets on Hugging Face. He contributed Turkish prompt-injection and data-leakage test cases to the OWASP GenAI Security Project's GenAI Data Security Initiative; the contribution was merged into the main repository. He took part in the Türkiye Siber Vatan Program, verified by the Republic of Türkiye Ministry of Industry and Technology, and holds a BlueDot Impact "Future of AI" completion certificate.
Let's talk about your organization's AI security needs
AltaySec provides LLM security consulting, AI penetration testing, red teaming, and training. We take KVKK and EU AI Act obligations into account in our work.