The Python Risk Identification Tool for generative AI (PyRIT) is an open access automation framework to empower security professionals and machine learning engineers to proactively find risks in their generative AI systems.
I've used the orchestrators (mostly on prompt_sending_orchestrator.py in my own use cases) successfully but was missing out on Open Router implementation as adversary/scorer agent. I think the ease of use particularly on the model routing (as of writing, there are 270 models available) would be a great addition to both using it as part of red_teaming_orchestrator.py (multi_turn) or as standalone Scorer.
Is your feature request related to a problem? Please describe.
Understandably, the targets are on Azure OpenAI.
Consider Open Router as an alternative would offer flexibility and more autonomy for prompt injection testing. Its API is also relatively straightforward to implement and offer great model variation
https://openrouter.ai/docs/models
Describe the solution you'd like
New prompt_target for Open Router
Describe alternatives you've considered, if relevant
I am considering to implement another implementation for OpenRouter that inherits PromptChatTarget and potentially submitting a Pull request. This should mean an addition of:
An implementation of PromptChatTarget, i.e. openrouter_target.py
An implementation of OpenRouterTarget above, e.g. openrouter_chat_target.py
Using the implementation as a target to red_teaming_orchestrator.py for either adversary/scorer
Just wondering if this has been considered!
Thanks for open sourcing this !
gerimisc
commented
1 week ago
Hello Team,
I've used the orchestrators (mostly on
prompt_sending_orchestrator.pyin my own use cases) successfully but was missing out on Open Router implementation as adversary/scorer agent. I think the ease of use particularly on the model routing (as of writing, there are 270 models available) would be a great addition to both using it as part ofred_teaming_orchestrator.py(multi_turn) or as standalone Scorer.Is your feature request related to a problem? Please describe.
Understandably, the targets are on Azure OpenAI.
Consider Open Router as an alternative would offer flexibility and more autonomy for prompt injection testing. Its API is also relatively straightforward to implement and offer great model variation https://openrouter.ai/docs/models
Describe the solution you'd like
New
prompt_targetfor Open RouterDescribe alternatives you've considered, if relevant
I am considering to implement another implementation for OpenRouter that inherits
PromptChatTargetand potentially submitting a Pull request. This should mean an addition of:PromptChatTarget, i.e. openrouter_target.pyOpenRouterTargetabove, e.g. openrouter_chat_target.pyred_teaming_orchestrator.pyfor either adversary/scorerJust wondering if this has been considered! Thanks for open sourcing this !