AI bezbednost Training in Belgrade

This instructor-led, live training in Belgrade (online or onsite) is aimed at beginner-level to intermediate-level IT professionals who wish to understand and implement AI TRiSM in their organizations.

By the end of this training, participants will be able to:

• Grasp the key concepts and importance of AI trust, risk, and security management.
• Identify and mitigate risks associated with AI systems.
• Implement security best practices for AI.
• Understand regulatory compliance and ethical considerations for AI.
• Develop strategies for effective AI governance and management.

This instructor-led training in Belgrade focuses on governance, identity management, and red-teaming for agentic AI systems. Advanced practitioners will learn to design secure deployments, enforce least-privilege access, and perform adversarial testing to mitigate real-world threats within production environments.

Ova obuka koju vodi trener, uživo u <lokalizaciji> (online ili licno), namenjena je profesionalcima iz oblasti AI-a i kibernetske bezbednosti srednjeg nivoa koji žele da razumeju i reše bezbednosne ranjivosti specifične za AI modele i sisteme, posebno u visoko regulisanim industrijama kao što su finanse, upravljanje podacima i konsalting.

Do kraja ove obuke, učesnici će moći:

• Da razumeju vrste adversarial napada na AI sisteme i metode za odbranu od njih.
• Da implementiraju tehnike jačanja modela radi obezbeđivanja mašinskih učenjima pipeline-ova.
• Da obezbede bezbednost i integritet podataka u modelima mašinskog učenja.
• Da se kretaju kroz regulatorne zahteve za usklađenošću vezane za bezbednost AI-a.

This instructor-led, live training in Belgrade (online or onsite) is designed for advanced-level security professionals and ML specialists who wish to simulate attacks on AI systems, uncover vulnerabilities, and enhance the robustness of deployed AI models.

By the end of this training, participants will be able to:

• Simulate real-world threats against machine learning models.
• Generate adversarial examples to test model robustness.
• Assess the attack surface of AI APIs and pipelines.
• Design red teaming strategies for AI deployment environments.

This instructor-led course in Belgrade enables advanced professionals to fortify TinyML pipelines on edge devices. It covers the implementation of privacy-preserving strategies, hardening models against adversarial risks, and adhering to best practices for secure data management in constrained environments.

This instructor-led, live training in Belgrade (online or onsite) is designed for intermediate-level engineers and security experts who aim to defend AI models deployed at the edge against threats such as tampering, data exposure, adversarial attacks, and physical breaches.

Upon completing this training, participants will be equipped to:

• Identify and evaluate security risks within edge AI deployments.
• Implement tamper-resistant measures and encrypted inference methods.
• Strengthen edge-deployed models and secure data pipelines.
• Deploy threat mitigation strategies tailored to embedded and constrained systems.

This instructor-led, live training in Belgrade (offered online or on-site) targets advanced professionals who aim to implement and evaluate techniques like federated learning, secure multiparty computation, homomorphic encryption, and differential privacy within practical machine learning pipelines.

By the end of this training, participants will be capable of:

• Understanding and comparing major privacy-preserving techniques in ML.
• Implementing federated learning systems with open-source frameworks.
• Applying differential privacy to ensure safe data sharing and model training.
• Using encryption and secure computation to protect model inputs and outputs.

This instructor-led live training in Belgrade (online or onsite) is designed for intermediate-level enterprise leaders seeking to understand how to responsibly govern and secure AI systems in compliance with emerging global frameworks such as the EU AI Act, GDPR, ISO/IEC 42001, and the U.S. Executive Order on AI.

Upon completing this training, participants will be able to:

• Grasp the legal, ethical, and regulatory risks associated with AI usage across various departments.
• Interpret and implement major AI governance frameworks, including the EU AI Act, NIST AI RMF, and ISO/IEC 42001.
• Establish security, auditing, and oversight policies for AI deployment in the enterprise.
• Develop procurement and usage guidelines for both third-party and in-house AI systems.

Ova obuka koju vodi instruktur u Belgrade (online ili uživo na licu mesta) namenjena je AI developerima, arhitektama i produkt menadžerima srednjeg do naprednog nivoa znanja, koji žele da identifikuju i ublaže rizike povezane sa aplikacijama pogonjenim LLM-ovima, uključujući ubacivanje promptova (prompt injection), curenje podataka i nefiltrirane izlaze, uz uvođenje bezbednosnih kontrole poput validacije ulaznih podataka, nadzora sa ljudskim učešćem (human-in-the-loop) i mehanizama za zaštitu izlaznih podataka.

Na kraju ove obuke, učesnici će moći da:

• razumeju osnovne ranjivosti sistema zasnovanih na LLM-ovima.
• primene principe bezbednog dizajna na arhitekturu LLM aplikacija.
• koriste alate poput Guardrails AI i LangChain za validaciju, filtriranje i bezbednost.
• integrišu tehnike poput sandboxing-a, red teaminga i pregleda sa ljudskim učešćem u produkcijske pipeline-ove.

Ovaj trening vođen od strane instruktora u Belgrade (online ili uživo) namenjen je profesionalcima srednjeg nivoa u mašinskom učenju i kibernetičkoj bezbednosti koji žele da razumeju i smanje rizik od novih pretnji prema AI modelima, koristeći konceptualne okvire i praktične mere zaštite, kao što su robustno treniranje i diferencijalna privatnost.

Konačno, učesnici ovog treninga biće u stanju:

• Da prepoznaju i klasifikuju pretnje specifične za AI, poput adversarialnih napada, inverzije i trovanja podacima.
• Da koriste alate kao što je Adversarial Robustness Toolbox (ART) za simulaciju napada i testiranje modela.
• Da primene praktične mere zaštite, uključujući adversarialno treniranje, ubacivanje šuma i tehnike čuvanja privatnosti.
• Da osmisle strategije evaluacije modela svesne pretnji u produkcionim okruženjima.

Ovo trenerom vođeno uživo obučavanje na <lok> (online ili licem u lice) namenjeno je profesionalcima iz IT bezbednosti, upravljanja rizicima i usklađenosti na početnom nivou koji žele da razumeju osnovne koncepte bezbednosti veštačke inteligencije, vektore pretnji i globalne okvire poput NIST AI RMF-a i ISO/IEC 42001.

Kroz obuku, učesnici će moći da:

• Razumeju specifične bezbednosne rizike koje uvode sistemi veštačke inteligencije.
• Identifikuju vektore pretnji poput adversarial napada, trovanja podacima i inverzije modela.
• Primenjuju osnovne modele upravljanja, kao što je NIST okvir za upravljanje rizicima veštačke inteligencije.
• Usklade upotrebu veštačke inteligencije sa nastajućim standardima, smernicama za usklađenost i etičkim principima.

Zasnivajući se na najnovijim smernicama projekta OWASP GenAI Security, učesnici će naučiti kako da identifikuju, procene i ublaže pretnje specifične za AI kroz praktične vežbe i stvarne scenarije.

This course offers a practical introduction to securing modern AI-powered applications, APIs, copilots, and autonomous agents. Participants will learn how AI security diverges from traditional web security, explore common AI-specific threats such as prompt injection, RAG poisoning, and agent abuse, and understand how to protect AI systems using layered defenses including WAFs, AI gateways, API security, and guardrails. Through hands-on labs and real-world examples, students gain the skills to identify AI attack patterns, secure LLM-based applications, and deploy effective runtime defenses for production environments.

This course teaches software developers how to build AI-powered applications securely by design. Participants learn how to protect chatbots, copilots, RAG pipelines, and AI agents against AI-specific threats such as prompt injection, data poisoning, tool abuse, secret leakage, and insecure model output. The course covers secure prompt design, RAG security, least-privilege access, guardrails, and red-team testing, helping developers build AI features that are secure, reliable, and resilient in real-world environments.

Ova trening obuka koju vodi instruktor, uživo u Belgrade (online ili licem u lice), namenjena je inženjerima bezbednosti i službenicima za usklađenost koji žele da ojačaju EXO implementacije, kontrolišu pristup modelima i upravljaju AI radnim opterećenjima koja se izvršavaju isključivo u lokalnim (on-premise) uslovima.

This instructor-led live training in Belgrade (online or onsite) targets security and ML engineers who need to identify, test, and defend against attacks on ML models and LLM-powered applications.