Originalarbeiterreichbar
Wallace et al. (OpenAI), The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions
Wallace und Mitautoren von Open
geprüft 10.11.2026
Worauf sich diese Seite beruft, wörtlich, abgerufen am 28.10.2026:
we argue that one of the primary vulnerabilities underlying these attacks is that LLMs often consider system prompts (e.g., text from an application developer) to be the same priority as text from untrusted users and third parties.
bestätigt 10.11.2026Today's LLMs are susceptible to prompt injections, jailbreaks, and other attacks that allow adversaries to overwrite a model's original instructions with their own malicious prompts.
bestätigt 10.11.2026