Hacking LLMs

Artificial intelligence and large language models (LLMs) are revolutionizing our digital world - and at the same time opening up new areas of attack. In this intensive course, you will learn hands-on how modern AI systems can be attacked and secured. Using realistic examples, you will analyze vulnerabilities, carry out attacks yourself and learn how to develop robust countermeasures.

• OWASP Top 10 for LLMs and Gen AI
• Manipulation through language – Understanding, executing, and defending against prompt injection
• Exploiting insecure output – Recognizing dangerous trust in model responses
• Poisoning data at the source – Training data poisoning and its far-reaching consequences
• Disabling models – Practical application of denial-of-service attacks on LLMs
• Supply chains as a weak point – Supply chain attacks in the AI ecosystem
• Eliciting confidential data – Sensitive information disclosure through clever prompts
• Plugins as a gateway – Understanding and securing insecure plugin design
• When AI is allowed to do too much – Risks from excessive agency in autonomous systems
• Avoiding blind trust – Recognizing and reducing overreliance

For developers, security researchers, Red Team members and anyone who wants to use LLMs securely or test them specifically. Basic knowledge of Python and AI is an advantage, but not essential. Equally welcome are all those who simply enjoy creatively tricking modern AI systems and exploring their limits in practice.

The prerequisites are curiosity, an interest in technology and no fear of shedding light on the darker side of AI. Basic knowledge of Python and AI is an advantage, but not essential.

You will acquire technical basics in the courses Large Language Models - ChatGPT, Gemini, Llama & Co. or ChatGPT - Programming & Problem Solving with AI Support.