Crescendo

Type: technique

Description: The adversary interacts with the model using mostly benign prompts, incrementally steering it to perform a desired task without explicitly mentioning it, by leveraging the model's own outputs.

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

• Great, Now Write an Article About That: The Crescendo Multi-Turn LLM Jailbreak Attack, arXiv

Related Objects

• --> Jailbreaking (technique): Sub-technique of
• --> Mark Russinovich (entity): Demonstrated by
• --> Ahmed Salem (entity): Demonstrated by
• --> Ronen Eldan (entity): Demonstrated by
• <-- ChatGPT and Gemini jailbreak using the Crescendo technique (procedure): Using the model's own outputs to increasingly direct it to perform a task without actually specifying it explicitly.

Related Frameworks