About
We build tools that stay safe when the model doesn’t.
Prompt injection isn’t a bug the next model fixes. The research that took it seriously moved the defence into the design, and we put that design in the hands of teams who build AI routines every day.
Principles
- Assume the model is fooled. Every guarantee we make holds when it is.
- Claims people can check. Keep examples and evidence separate. Say when existing tools are sufficient, show what Cresec adds, and back each claim with something you can check.
- Say when you don’t need us. If a routine never sends anything out, we’ll tell you.
The team
- Jonathan Rapoport, founder, with a background in cybersecurity, AI infrastructure and AI model training · LinkedIn profile of Jonathan Rapoport (opens in a new tab)