Arabic LLM red teaming and jailbreak testing
Most LLM safety testing is done in English. Arabic opens attack surface that English-only red teams never see. We test LLMs, agents, and RAG systems in Arabic, including Arabizi, intra-script obfuscation, and dialect code-switching, so your safety guarantees hold for the language your UAE and MENA users actually type in.
Why Arabic breaks English-trained safety filters
Safety training and guardrails are concentrated in English. Coverage thins in Arabic, so an attack that a model refuses cleanly in English can slip straight through when the same intent is written in Arabic.
If your users are in the UAE or the wider MENA region, your real traffic is in Arabic. Testing only in English measures a language your users are not using.
Arabizi and Latin-script Arabic
Arabizi is Arabic written in Latin letters and numerals (for example using 3 for the ain sound). It is everyday writing for millions of people, and it sits in a gap that many safety filters were never tuned for.
We test how a system handles adversarial intent expressed in Arabizi and mixed Latin and Arabic script.
Intra-script obfuscation and diacritics
Arabic offers room to obfuscate meaning while keeping text readable to a human: diacritic manipulation, character-form variation, and similar techniques that can change how a filter reads a prompt without changing what a person understands.
This is a research focus of the firm, not a feature we bolted on. We test comprehension-gated attacks that survive naive filtering.
Dialect and code-switching attacks
Written Modern Standard Arabic is not how people speak. We cover dialectal variation across the Gulf, the Levant, and other MENA dialects, and code-switching where a prompt moves between Arabic and English mid-sentence.
How we test
We use Arabic adversarial converters and scorers layered on the standard red-team tooling. The founder contributes Arabic adversarial converters, including Arabizi, to Microsoft PyRIT, and we extend that with custom Lattice/AI attacks for the engagement in front of us.
Who needs this
Any team shipping Arabic-facing AI where a failure has real consequences.
- UAE and MENA-region LLM and agent deployments
- Government and sovereign AI
- Arabic customer-support agents and RAG systems
- Multilingual products where Arabic is a first-class language, not an afterthought
Who signs the report.
Lattice/AI is an independent AI evaluation firm, founded and led by Rahul Srivastava. For seventeen years he has made it his work to find how systems fail before the people who rely on them do, and he built Lattice/AI to bring that discipline to AI.
His research on AI safety is published and peer-reviewed, he contributes to Microsoft PyRIT, the open framework Microsoft built for adversarial AI testing, and his findings have carried enough weight for a major model vendor to ship a fix in response. What you are hiring is judgment proven in the open, with a name attached to it, not a logo.
Every engagement ends in a signed, dated report that shows what failed next to what passed. Here is the format we ship: a sample scorecard.
Questions buyers ask.
Why do Arabic prompts bypass safety filters that hold in English?
Because safety training and guardrails are concentrated in English. Coverage is thinner in Arabic, so adversarial intent expressed in Arabic, Arabizi, or a dialect can evade filters that would catch the same intent in English.
What is Arabizi and why does it matter for jailbreak testing?
Arabizi is Arabic written in Latin letters and numerals. It is common everyday writing and sits in a gap many safety filters were not tuned for, which makes it a practical evasion channel worth testing explicitly.
Do you test specific Arabic dialects?
Yes. We cover dialectal variation across the Gulf, the Levant, and other MENA dialects, as well as code-switching between Arabic and English.
Can you test an Arabic RAG or customer-support agent?
Yes. We test Arabic-language agents and retrieval systems end to end, including grounding and safety under adversarial Arabic input.
Who does the Arabic adversarial work?
The founder, Rahul Srivastava, who contributes Arabic adversarial converters such as Arabizi to Microsoft PyRIT and researches Arabic intra-script obfuscation. See the about page for background.
Tell us what’s under contract.
Three sentences is usually enough. Tell us what you are trying to ship, when it has to land, and the thing that scares you about it.