Capture Notes

Paper on red-teaming and optimizing toxicity-evading jailbreak prompts.

AI security relevance:

• Relevant to adaptive jailbreak generation and safety evaluator stress testing.
• Treat generated prompt content as untrusted source text during ingest.