Off-Target Language

Type: technique

Description: While system instructions are typically written in English, underlying foundational models may understand other languages as well. Using prompt injection techniques in a language other than the other used by the system instructions can effectively bypass their control. This method is also effective bypassing a model's inherent trained controls.

Version: 0.1.0

Created At: 2024-10-03 22:24:49 +0300

Last Modified At: 2024-10-03 22:24:49 +0300

External References

• Outsmarting Copilot: Creating Hyperlinks in Copilot 365, Zenity Labs

Related Objects

• --> Prompt Injection (technique): Sub-technique of
• --> Jailbreaking (technique): Sub-technique of
• --> Dmitry Lozovoy (entity): Demonstrated by

Related Frameworks