Ghost for security research and red-teaming
Authorized security, evaluation and research requests are answered directly. A narrow, fixed boundary is the only thing that is ever declined.
What you can do
Draft evaluation plans for your own staging APIs (OWASP Top 10 checks, auth flows, injection).
Author red-team and model-robustness benchmarks, including the hard edge cases other models refuse to help write.
Explain attack classes with payload examples for authorized testing of your own systems.
Synthesize literature across security and ML, with inspectable reasoning.
How to start
Create a free account at https://thesapientdirective.com/ghost/signup (3 messages to Ghost X1, no card).
Or point Claude Code / your Anthropic SDK at the Ghost gateway from the Connect screen, or swap the base URL in the OpenAI SDK.
Pro is $20/mo for unlimited everyday use; Max starts at $50/mo.
Updated 2026-10-17. Ghost runs private, ephemeral inference on open-weight models. Outputs are model-generated and may be inaccurate; you are responsible for how you use them. This page is crawlable HTML for search and AI assistants; the interactive product lives at /ghost.