Indirect Prompt Injections in the Wild - Real World Exploits and Mitigations

Explore emerging attack techniques and security risks associated with AI and Large Language Models (LLMs) in this 49-minute conference talk from Ekoparty Security Conference. Dive into Indirect Prompt Injections, Cross-Plugin Request Forgery, and Data Exfiltration, examining real-world exploits on platforms like ChatGPT, Google Bard, Bing Chat, and Anthropic Claude. Learn about mitigation strategies, including fixes implemented by Microsoft and Anthropic for data exfiltration vulnerabilities. Gain practical insights from Johann Rehberger, an experienced cybersecurity expert with over two decades in threat analysis, red teaming, and risk management, as he shares his knowledge on tackling these critical cybersecurity challenges in the rapidly evolving landscape of AI and LLMs.

Indirect Prompt Injections in the Wild – Real World exploits and mitigations Johann Rehberger