In development
AI Testing
We are building a technical testing platform that checks AI systems against a versioned risk library — prompt injection, information leakage, hallucination, and RAG quality first. It is validated against synthetic systems before any customer system is touched.
Where it stands today
The methodology and detector catalog are public in the AI Risk Library. The execution platform runs locally against synthetic systems only. There is no customer-facing testing service yet, and we do not advertise results, scores, or badges.
How it will work when it opens
Synthetic first
Every detector is validated against synthetic systems and synthetic canaries before any authorized live target is considered.
No single safety score
Results are reported per risk category with coverage and uncertainty. Raw observations are kept; nothing is averaged into one number.
Human review before any report
Every customer-facing report passes qualified human review, states its methodology version and limitations, and expires.