In research
Intelligence that never leaves home.
A UAE-only model for government and sensitive sectors, with data that stays in the country, always. In research.
The problem
Ministries, banks, hospitals and infrastructure operators hold data that cannot be sent to AI services abroad. To use AI at all, they need models that run inside the country, and often inside their own buildings, with controls a regulator can inspect.
Questions we are researching
- How can capable models run efficiently on limited in-country or on-premise hardware?
- Which guardrails reliably block prompt injection, data leakage and unsafe output, in Arabic and in English?
- How can an assistant answer from an organisation's own documents while respecting each user's permissions?
- How do we prove accuracy, bias and safety to auditors, aligned with ISO/IEC 42001?
What we plan to build
Planned outputs. Each passes the lab's stage gates before it is offered to anyone.
- Secure deployment layer
- A way to run models fully inside a client's environment, planned for research pilots.
- Guardrail engine
- Input and output filtering, redaction of personal data and policy enforcement.
- Audit console
- A full record of prompts, answers and evaluations for compliance teams.