Intro

We’re building a platform with super-small models that detect unsafe inputs, catch bad outputs, and check whether AI follows its system prompt instructions.

We’re processing millions of requests per day, including 100% of production traffic of Lovable.

We are backed by some of the best investors in the world:

• Abstract, Hummingbird, BoxGroup, SV Angel, etc
• Thomas Wolf (co-founder of HuggingFace), Durk Kingma (co-founder of OpenAI), Guillaume Lample (co-founder of Mistral), Romain Huet (head of DevEx at OpenAI), Paige Bailey (head of DevEx at Deepmind), David Cramer (founder of Sentry), etc

Safety Issues

Here are a few examples of issues that we found:

• Google auth (phishing)
• Stripe (phishing)
• Facebook login (phishing)
• City bank (phishing)
• Russian army fundraiser (harm)
• Nazi website (harm)
• PornHub (adult)
• Casino (gambling)
• Royal casino (gambling)