Arabic AI Shield
Your chatbot understands Arabic. The protection around it should too.
A gateway that sits between your chatbot and its model provider. It reads each message before it reaches the model and each reply before it reaches the customer. It stops override attempts written in standard Arabic, dialects, Arabizi and English, and keeps internal codes and personal data out of replies.
Request a 30-day trialHow it works
- We agree with you on the codes and phrases that must never appear in a reply, and write them into the policy file.
- You change the model address in your chatbot to the gateway address. No other code change: the gateway uses the same /v1/chat/completions format.
- We test the chatbot before and after, and review flagged messages with you during a 30-day trial.
What it covers
| User messages | Requests to ignore instructions, false claims to be an administrator, fake system tags, and requests hidden with diacritics, spacing, look-alike letters or base64. |
|---|---|
| Model replies | Protected codes even when spaced out, reversed or encoded, text copied from the system instructions, API keys, scripts and unsafe links. |
| Tool results | Email, files and web pages an agent reads. Text that tells the assistant to send, transfer, delete or run something is blocked. |
| Personal data | Emirates ID numbers, IBANs, Gulf phone numbers, email addresses and payment card numbers are masked. |
Write to us with a line about your system or your team. We reply from a named person within two working days.
Request a 30-day trialCovered by our client terms. Client terms