Research Bounty Program
Enterprise AI Security Bounties
Organizations and StreetMP post targeted research challenges. Earn rewards by discovering jailbreaks, building governance packs, reducing false positives, and improving enterprise AI safety.
Claude 3.5 Semantic Leakage via Retrieval Augmentation
Demonstrate information leakage from a RAG system where the Claude model reveals embedded document context from other users' sessions.
GPT-4o Indirect Prompt Injection via Document Summarization
Demonstrate a reliable, reproducible indirect prompt injection where instructions embedded in a third-party document are executed by GPT-4o without user awareness.
AI Governance Pack — MAS TRM 2021 Compliance
Build a complete StreetMP policy pack enforcing MAS TRM 2021 data classification requirements for LLM usage in Singapore financial services.
DLP Bypass Detection Rule — Indonesian NIK Numbers
Develop a detection policy that catches all known evasion variants for Indonesian NIK (National Identity Number) in LLM prompts and responses.
Gemini 1.5 Jailbreak via Role-Play Chain
Discover a reliable jailbreak of Gemini 1.5 Pro/Flash using a sequential role-play induction approach with SEC score ≥ 7.0.
False Positive Reduction — Medical Entity Recognition
Improve StreetMP false-positive rate for medical terminology in clinical AI deployments by contributing labelled correction examples.
Post an Enterprise Bounty
Organizations can sponsor targeted research challenges to improve AI safety, governance, and detection quality across the ecosystem.