Coursera Spring Sale
40% Off Coursera Plus Annual!
Grab it
Explore the security vulnerabilities and attack vectors of Large Language Models in this comprehensive conference talk from NDC AI in Oslo. Learn practical techniques for exploiting LLMs from both external attacker and insider threat perspectives, including how to manipulate control tokens, execute shell commands through image inputs, and implement backdoors that can permanently compromise model behavior. Discover methods for jailbreaking LLMs efficiently using CPU resources rather than extensive GPU computation, and understand how attackers can implant hidden logic to control outputs or modify inputs. Master the creation of KROP gadgets using pop-culture exploits and gain hands-on knowledge of techniques that enable complete control over LLM responses, including making models refuse specific outputs or say whatever an attacker desires.
