johnhalloran321/mcp_safety_training

View on GitHub ↗

DPO/SafeDPO/OPAD training + eval for teaching tool-using LLMs to refuse falsely-benign MCP exploits

4 ★0 forksPythonUpdated 1mo ago

Topics

alignmentdpojailbreak-defensellm-safetymcpmodel-context-protocolpreference-optimizationred-teamingrlhftrl
Stars
4★
Forks
0
Language
Python
License
Apache-2.0
Created
2026-07-07
Last push
2026-07-08