Update API so that users can submit prompts with the "paraphrase" flag. The submitted prompt will be paraphrased by an LLM and returned with the normal scan results.
Idea is that if the prompt contains an injection, paraphrasing while retaining the overall meaning may break injection strings.
Probably something like this
POST /analyze/prompt?paraphrase=true
Update API so that users can submit prompts with the "paraphrase" flag. The submitted prompt will be paraphrased by an LLM and returned with the normal scan results.
Idea is that if the prompt contains an injection, paraphrasing while retaining the overall meaning may break injection strings.
Probably something like this POST /analyze/prompt?paraphrase=true