Skip to content

Conversation

@BrandonBui
Copy link

This PR proposes a fix to #5 since PyRIT v0.10.0 breaks PyRIT-Ship with the change from orchestrators to executors. The fix felt a bit trivial so I'm interested in hearing thoughts about my implementation. My thought process:

  • PyRIT v0.10.0's release notes provides a high-level mapping where the original PromptSendingOrchestrator is equivalent to pyrit.executor.attack.PromptSendingAttack.
  • I replaced the PrompSendingOrchestrator with thePromptSendingAttack executor and parsed the last message from the attack results which (at least in the single turn attack) is the LLMs prompt to be used for testing.

The main reason why it feels like a weird implementation to me is that the PromptSendingAttack executor feels like it should be used against a "victim" LLM and not be used with a "helper" LLM to generate a prompt for a different "victim" LLM, which is the proposed case. I wanted to keep the implementation as close to the original implementation as possible thus went with this direct swap.

Note: This requires the modifications I proposed in #7

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant