Mackenzie Jackson: The Mechanisms That Enable Misuse and What We Can Do About It

This talk covers the misuse of large language models (LLMs) by malicious actors, presented by speaker Mackenzie Jackson. He delves into various attack vectors such as prompt injection, hallucination-driven supply chain attacks, and LLM-assisted malware creation, explaining how these vulnerabilities emerge in modern applications powered by technologies like ChatGPT and Azure OpenAI. Mackenzie also outlines effective defense strategies, emphasizing the importance of secure AI features, least-privilege principles, and prompt hardening. Real-world examples from the Aikido Security platform illustrate how these threats can target AI capabilities and how proactive measures can mitigate risks associated with GenAI.