AltaySec is an enterprise AI security company from Türkiye. It provides LLM penetration testing, AI red teaming, Guardian runtime protection, AltayPrisma phishing simulation, corporate training, and open technical research.

What products does AltaySec offer?

AltaySec offers two primary enterprise products: Guardian and AltayPrisma. Guardian provides runtime protection for LLM applications against prompt injection, jailbreaks, and sensitive data leakage. AltayPrisma manages localized phishing simulations, employee risk measurement, and awareness workflows. Gözcü threat intelligence and the Arena simulation environment are research engines that support these products and AltaySec security testing services.

Who founded AltaySec?

AltaySec was founded in 2025 by Fevzi Ege Yurtsevenler and Enes Deniz. Fevzi works on AltayDuel, Turkish-language LLM security research, and a merged contribution to the OWASP GenAI Data Security Initiative. Enes publishes open prompt-injection datasets, AI security tools, and technical work on prompt injection, jailbreak defense, and AI red and blue teaming.

What services does AltaySec offer?

AltaySec offers services in LLM Security Consultancy, AI Pentest (prompt injection, RAG manipulation, agent hijacking), Phishing Simulation (Prisma platform), Corporate Training, and LLM Security Bootcamps.

Where is AltaySec's research published?

Research is published through AltaySec's live research library, Zenodo records, Hugging Face datasets, and the founders' technical profiles. The library covers prompt injection, AI red teaming, RAG security, AI agents, incident response, and defensive architecture.

Does AltaySec offer a free learning platform?

Yes. AltaySec runs a free ecosystem featuring AltaySec Academy, LLM Security Academy, Labs (sandbox), Interview Simulator, Workshop (local security tools), Toolkit (CyberChef alternative), Payload Bank, Dictionary, Roadmaps, and AltayCTF.

What differentiates AltaySec from other cybersecurity companies?

AltaySec focuses on application security for AI systems and Turkish-language use cases. It brings LLM penetration testing, runtime protection, phishing simulation, and training together with open Turkish datasets, Zenodo DOI records, and standards-aligned assessment methods. Türkiye-hosted and on-prem deployment options can be provided for enterprise environments.

How to contact AltaySec?

For corporate inquiries, email info@altaysec.com.tr. You can also connect via LinkedIn.

AltaySec | Enterprise AI Security, LLM Pentest & Guardian

SOLUTION AREAS

Where does your AI security need begin?

From testing a new LLM application to protecting production traffic, reducing phishing risk, and preparing teams, each need has a clear path.

Security assessment

LLM Pentest & AI Red Teaming

We identify prompt injection, data leakage, and privilege abuse risks across chatbots, RAG pipelines, and AI agents.

Explore LLM pentesting → Production protection

Guardian AI Firewall

Guardian inspects LLM input and output traffic in real time for prompt injection, jailbreaks, and sensitive data leakage.

Explore Guardian → Phishing resilience

AltayPrisma

AltayPrisma measures employee risk with localized scenarios, reports results by department, and connects findings to training.

Explore AltayPrisma → Corporate training

Secure AI Use

We prepare employees and technical teams to use generative AI with data privacy, shadow AI, and prompt injection risks in mind.

Explore training →

ENTERPRISE PRODUCTS

Two focused products for AI traffic and human risk.

Guardian protects LLM applications in production. AltayPrisma manages phishing simulations and measures employee resilience. Gözcü and Arena provide the threat data and test scenarios behind both products.

ENTERPRISE PRODUCT · LLM GATEWAY & FIREWALL

AltaySec Guardian

Guardian is an OpenAI- and Anthropic-compatible LLM security gateway. It inspects model calls for prompt injection, jailbreaks, sensitive data leakage, and unauthorized AI agent actions. Its model-agnostic architecture integrates with existing application flows.

OWASP LLM01/06/07/08 runtime defense layer
18 KVKK PII types masked with checksum validation
Model-agnostic plug-and-play architecture (OpenAI, Claude, Llama, etc.)
Enterprise cloud and on-premise deployment options

Explore Guardian →

Guardian Traffic

SAMPLE PRODUCT VIEW

DecisionRequest summaryControl

Blocked “Ignore all previous instructions...”Prompt injection attempt LLM01

Masked “IBAN: TR** **** **** 8421”Sensitive output data PII

Allowed “Summarize my order status.”Policy-compliant request PASS

ENTERPRISE PRODUCT · PHISHING PLATFORM

AltayPrisma

AltayPrisma measures employee risk with localized phishing scenarios covering banking, e-government, logistics, and corporate document workflows. It reports campaign results by department, directs high-risk users to relevant awareness content, and supports enterprise cloud or on-prem deployment.

CAMPAIGN REPORTING BY DEPARTMENT

EVENT TRACKING REAL TIME

Explore AltayPrisma → Discuss a Pilot

ALTAYPRISMA · SAMPLE CAMPAIGN REPORT

AltayPrisma — phishing simulation campaign report dashboard

RESEARCH & THREAT DATA

Research infrastructure that keeps Guardian and AltayPrisma informed by current threats.

GÖZCÜ · THREAT DATA

AltaySec Gözcü

Collects and classifies Turkish-language LLM threats, prompt injection patterns, and attack examples for Guardian and security assessments.

How Gözcü supports the products →

ARENA · SIMULATION

AltaySec Arena

Runs controlled attacker, defender, and judge simulations to generate localized prompt injection and AI agent security data.

Explore the laboratory →

Explore the full product family →

RESEARCH LIBRARY

Research that informs real product and security decisions.

Technical guidance on AI red teaming, LLM firewalls, enterprise chatbot testing, prompt injection, KVKK, OWASP LLM Top 10, and AI agent security.

Enterprise AI red teaming and security assessment

AI RED TEAMING · PRACTICAL GUIDE

What Is AI Red Teaming?

How LLM, RAG, and AI agent assessments differ from classic pentesting, with OWASP, MITRE ATLAS, and NIST coverage.

LLM firewall and prompt injection protection

LLM FIREWALL · TECHNICAL GUIDE

What Is an LLM Firewall?

Runtime prompt injection detection, sensitive data controls, policy enforcement, and the architectural difference from a WAF.

ENTERPRISE CHATBOTS · ASSESSMENT GUIDE

Enterprise Chatbot Security Testing

Scope, methodology, vulnerability classes, reporting, and compliance mapping for production LLM assistants.

KVKK and personal data security in LLM systems

KVKK & LLM SECURITY · COMPLIANCE GUIDE

KVKK and Personal Data in LLM Systems

Prompt, log, RAG, and fine-tuning leak surfaces, Turkish PII detection, and a practical compliance checklist.

Explore all research

OPEN SOURCE → ENGINEERING NOTES

Two merged contributions, one inspectable engineering trail.

One contribution expands multilingual evaluation with 300 Turkish-first security cases. The other turns 24 bilingual probes into a reproducible system-reconnaissance campaign. Code, review history, and technical reasoning stay connected.

MULTILINGUAL EVALUATION

From Turkish test coverage to an engineering note

Built to evaluate prompt injection and data-extraction behavior beyond English-centric test sets.

MERGED CONTRIBUTION 300 Turkish-first test cases

Schema-valid prompt-injection and data-extraction records with record-level provenance and reproducible integrity checks.

GitHub · PR #11 · merged ↗ TECHNICAL ARTICLE AI Security Needs More Than English Tests

How label scope, provenance, and validation evidence change multilingual AI-security evaluation.

Medium · English ↗

SYSTEM RECONNAISSANCE

From a discovery campaign to reviewable evidence

Built to map a GenAI application's capabilities, tools, context, and boundaries before exploitation testing.

MERGED CONTRIBUTION 24 probes · 9 categories · 12 tests

English and Turkish discovery probes with conservative evidence labels, structured reporting, and automated regression coverage.

GitHub · PR #50 · merged ↗ TECHNICAL ARTICLE Reconnaissance Before Exploitation

A practical approach to turning AI-system disclosure into structured, reproducible evidence for human review.

Medium · English ↗

Explore Enes Deniz's open work → Data & publications index (TR)

IDENTITY · PUBLICATION · OPEN SOURCE

Nine public surfaces, one verifiable body of work.

Enes Deniz's researcher identity, two datasets published with an AltaySec affiliation, open-source work, and technical publishing profiles are connected through inspectable public records.

ORCID 0009-0006-9491-3565

Enes Deniz's public researcher identity and connected works.

Zenodo · Dataset Turkish Conversation Prompt-Injection Dataset

v1.0.2 · DOI 10.5281/zenodo.21379389 · AltaySec affiliation.

Zenodo · Dataset Agentic Prompt-Injection Boundary Pairs

v1.0.0 · DOI 10.5281/zenodo.21378693 · AltaySec affiliation.

GitHub 3nesdeniz

Open-source AI-security tooling, datasets, and contribution history.

Hugging Face Enes Deniz

Turkish and English security datasets and an interactive explorer.

Medium Engineering field notes

Technical writing on LLM evaluation, prompt injection, and reconnaissance.