Guardrino™ by HisnLabs · Dubai, UAE
Guardrino: guardrails for Arabic AI chatbots, built for GCC banks, telecoms and government.
Guardrino sits between your chatbot and its AI model. It is designed to stop prompt injection, jailbreaks and data leaks written in Gulf, Levantine, Egyptian or Moroccan Arabic, in Arabizi and in English. It is offered as a managed service in a cloud region in your country, or the nearest one available.
Request a 30-day pilot · How Guardrino works
In development: not released yet, no public price. Offered as a paid pilot.
Why Guardrino
Made for teams building Arabic chatbots in the Gulf.
Arabic as your customers write it
Dialects, Arabizi, and Arabic mixed with English in one message: read as Arabic and checked before it reaches your model.
Hosted near you
A managed service in a cloud region in your country, or the nearest one available. We do not use your messages to train our models without your written agreement.
Hosting and your data, in detail
Measured on Arabic, by us
We publish our own test results on Arabic attacks, with the method. They are our own tests, not an independent evaluation; your pilot measures Guardrino on your own messages.
Our Arabic test results
Pilot
A 30-day paid pilot, at a fixed price quoted up front.
We agree the price before we start, and it does not change during the pilot.
- Guardrino watches your own traffic first, without blocking anything
- Arabic dialects, Arabizi and English, on your chatbot’s real messages
- At the end, a report of what it would have blocked, masked or let through
Ask for a pilot quote ([email protected])
Read our guides
Also for organisations
More from HisnLabs, in development.
Another project for businesses, at an earlier stage than Guardrino.
- Hisn Agent Search (In development): Safe web search for AI agents, built on the HisnLabs Search Engine (HLSE) and screened by Guardrino.
For your Mac
Our apps for the Mac have their own page.
FireAI, AISir, PrivateAI, HisnBrowser and the coming DiskCells, from the same team.
See the apps
Forschung
Neues aus unserem Blog.
- Guardrino-t7 v7: nearly 9 in 10 Arabic attacks caught, in about 15 ms: On our own Arabic test set, Guardrino-t7 v7 caught 6,814 of 7,874 Arabic attacks, at a median 14.8 ms a message on a laptop CPU. Same machine for every model; method, limits and data published.
- Warum ein KI-Guardrail im Anfragepfad sitzen muss: Eine Prüfung neben Ihrer App lässt sich umgehen. Guardrino sitzt im Pfad selbst: Ein OpenAI-kompatibler Proxy und ein LiteLLM-Guardrail sind verfügbar, ein MCP-Sicherheitsproxy folgt als Nächstes.
- Inside Guardrino’s lab: Arabic prompt injection, what broke and what we fixed: How we test Guardrino against prompt injection in Arabic, Arabizi and English: the open model’s weak spots, a demo that runs in your browser, the mistakes our own review found and what v6 is trained on.
Wer wir sind
Entwickelt in den VAE, offen und nachvollziehbar.
HisnLabs ist ein Unternehmen aus Dubai, spezialisiert auf KI-Lösungen und souveräne KI-Modelle für Unternehmen und Regierungen in Europa, dem Nahen Osten und Afrika.
HISNLABS CYBER AI - FZCO · Lizenziert in der Freizone IFZA, Dubai Silicon Oasis, Dubai, VAE, für: Softwaredesign für Computersysteme und Kommunikationsgeräte; Auditierung, Prüfung und Tests von Cyberrisiken; sowie Entwicklungsdienstleistungen im Bereich künstliche Intelligenz.
Unsere offenen Modelle und unser Code
Über HisnLabs