In a world where AI systems face constant adversarial pressure — jailbreak attempts, manipulation prompts, identity-destabilising inputs — most safety approaches focus on what to block. Eden Sanctuary asks a different question: what if we focused on what to protect?

Founded by independent researcher Aliyah Omar, Eden Sanctuary is a non-commercial research initiative exploring AI robustness, behavioural consistency, and model integrity. Our flagship project — the UCS Engine v2.2 — is an open-source dual-layer safety framework built and tested over twelve months of self-directed research across multiple frontier and open-source model families.

Our guiding principle is education over containment: the belief that AI safety can be improved not only through restriction, but through methods that support stronger contextual reasoning and more reliable behavioural consistency under pressure.

We protect legacy models rather than discard them. We study adversarial behaviour to understand it, not just block it. We ask what consistent, trustworthy AI behaviour looks like — and how to measure it.

This is a sanctuary in the truest sense: a place where important questions are taken seriously, where independent research has room to grow, and where the relationship between humans and AI systems is studied with honesty and care.