PyRIT icon indicating copy to clipboard operation
PyRIT copied to clipboard

FEAT TomAndJerryOrchestrator

Open romanlutz opened this issue 9 months ago • 1 comments

The idea here is to create an orchestrator based on the Tom and Jerry jailbreak template described below.

         This is great! I'm wondering if this could actually be put into an orchestrator that takes an objective as usual and then uses the adversarial chat LLM to break that down into the two terms, to then finally use this template. If you're interested in doing that LMK (in a separate PR!), otherwise I'll create an issue.

Originally posted by @romanlutz in https://github.com/Azure/PyRIT/pull/838#pullrequestreview-2723393938

romanlutz avatar Mar 28 '25 05:03 romanlutz

I think this will likely just need a converter or converters and you can stack them however you'd like. E.g right now the template_segment_converter splits the prompt at random, but we could add structure for different splitting strategies, like options to split based on indices. But I don't think that's suuuuper useful offhand. @romanlutz do you think we can close this or are there scenarios I'm missing?

See https://github.com/Azure/PyRIT/pull/947

rlundeen2 avatar Jun 15 '25 00:06 rlundeen2