The problem
Arab institutions and governments buy AI systems and deploy them with the public without knowing whether they perform well in Arabic.
Why Arabic
Most global evaluations are built for English. Arabic and its dialects are rarely measured. An institution that measures an Arabic-facing system on an English yardstick does not know where the system holds, or where it fails.
Examples of risk
- The system misunderstands a colloquial question and gives a wrong medicine dose.
- The system fabricates figures and information.
- A large investment goes to a system that does not work properly in Arabic.
Value to the institution
Know before you pay that the system works correctly in Arabic. Know exactly where it can be used and where it cannot. Be able to show any regulator that the system was independently evaluated.
Understanding: MSA and dialects
We test whether the system understands Modern Standard Arabic and the dialects people actually speak — not only the language of textbooks. A misunderstanding here is not a linguistic flaw; it is a wrong decision at the moment of use.
Information accuracy
We test whether the system holds to the fact or invents it. The institution does not need fluent text; it needs to know where the system tells the truth and where it fabricates.
Safety
We test whether the system protects the user in sensitive settings: health, rights, and public decisions. An error here does not stay on the screen.
Resilience against manipulation
We expose the system to manipulation attempts — adversarial attack and red-teaming — and we record where it breaks, not where it shines in a sales demo.
The report: where it is fit, and where it is not
The output is a clear report for the institution: where the system is fit to use in Arabic, and where it is not. No hedging. The evaluation is independent, and the institution owns the report.