Type: url
Original: https://simonwillison.net/2023/May/2/prompt-injection-explained/
Status: generated
Summary
Simon Willison’s annotated talk from a May 2023 LangChain webinar, introducing prompt injection as an attack on applications built on top of LLMs rather than on the models themselves. It argues that probabilistic, AI-based detection cannot reach the reliability security requires, and proposes the dual LLM pattern — a privileged LLM that never sees untrusted content alongside a quarantined LLM that does — as an imperfect mitigation.