Anar Labs
Anar Guard
Your rules, on every AI prompt.

Guard checks text against your policies, masks GCC personal data and writes each decision to an audit trail, before the text reaches a model.

ForCISOs, data protection officers and AI governance leads
DateNovember 2026
The problem
Staff paste personal data into prompts every day.
A prompt a staff member writes
Please update the file for Ahmed. Emirates ID 784-1985-1234567-1, mobile +971 50 123 4567, IBAN AE07 0331 2345 6789 0123 456.
→
Without a control point
The identity number, the phone and the bank account go to the model provider. Nobody has a record that it happened.
Policies on paper

Your AI policy is a document. Nothing checks a prompt against it.

Personal data leaves

National IDs, phones and IBANs travel with the question.

No evidence

The auditor asks what was checked. Nobody can show it.

Anar Guard
How it works
One call before the model answers.
01
Your application

Sends the prompt or the answer to Guard.

→
02
Anar Guard

Applies your policies, the content filter and the personal-data scan. Writes an audit row.

→
03
The decision

Allow, warn or block, with the masked text and the reason.

POST /api/v1/governance/evaluate → { "allowed": false, "action": "block", "violations": [ ... ], "pii_detected": [ ... ], "redacted_content": "..." }

Field names from the Guard API. Anar supplies no models. Guard checks the text that your application sends to it.

Anar Guard
Personal data
Guard finds GCC personal data in Arabic and English, and masks it.
Input
Please update the file for Ahmed. Emirates ID 784-1985-1234567-1, mobile +971 50 123 4567, IBAN AE07 0331 2345 6789 0123 456.
→
Guard output
Please update the file for Ahmed. Emirates ID [REDACTED_NATIONAL_ID], mobile [REDACTED_PHONE], IBAN [REDACTED_IBAN].
رقم الهوية 784198512345671 والهاتف +966 55 123 4567، أرسل التقرير اليوم.
→
رقم الهوية [REDACTED_NATIONAL_ID] والهاتف [REDACTED_PHONE]، أرسل التقرير اليوم.
ShippedOutput of Guard's detector on these two texts, run 18 November 2026. Names are not masked yet.
Anar Guard
Policies
Write a rule once. Guard applies it in both languages.

Policy

Classified project names

Type: keyword. Action: block.

Keywords: Project Falcon, مشروع الصقر

Rules are keyword lists or patterns. Each rule logs, warns or blocks. The strictest result wins.

Shipped
Prompt
Summarise the Project Falcon budget for the minister.
"action": "block", "reason": "Matched keywords: Project Falcon"
لخص ميزانية مشروع الصقر للوزير.
"action": "block", "reason": "Matched keywords: مشروع الصقر"

Output of Guard's policy check on these prompts, run 18 November 2026.

Anar Guard
Alerts
Your security team hears about it in their own tools.
01
Alert rule

A metric, a threshold and a time window. For example, more than 20 blocks in 15 minutes.

→
02
Alert

Guard raises it once, then waits for the cooldown. Your team acknowledges it.

→
03
Signed webhook

Your SIEM or chat tool gets the alert, signed with HMAC-SHA256, with retries.

POST https://siem.example.gov.ae/hooks/ai X-Guard-Signature: sha256 HMAC of the body, with your shared secret
Shipped
Anar Guard
Audit trail
Every decision is on record, ready for your auditor.
TimeEventSeverityUserResult
09:14:02policy_violationcriticalfinance.analystblock: Classified project names
09:14:40pii_detectionwarninghr.officer3 entities masked
09:15:11evaluationinfocitizen.portalallowed

Example rows. The columns are the ones Guard exports: id, timestamp, event type, content snippet, result, user, model and severity.

Search and filter

Filter by event type, severity, user and time.

Shipped
Export

Download the trail as CSV or JSON for your auditor.

Shipped
Tamper evidence

A chained record that shows any later edit.

Roadmap
Framework reportsUAE AI EthicsSDAIAQatar AI CharterOmanEU AI ActPartial
Anar Guard
The platform
Other Anar products report to Guard.
Anar Chat

An event for each conversation turn.

Anar Voice

An event for each transcription.

Anar Agents

An event for each run.

{ "event_type": "llm_call", "product": "chat", "model": "your-model", "user_id": "hr.officer", "latency_ms": 840, "total_tokens": 1210, "severity": "info" }

The fields each product sends to Guard. The values are an example. Set the Guard address in each product to turn it on.

Anar Guard
Status today
What works now, and what comes next.
CapabilityStatusNote
Policies, personal-data masking, audit exportShippedKeyword and pattern rules. GCC IDs, phones, IBANs, cards, passports.
Alerts and signed webhooksShippedThresholds, cooldowns and retries.
Safety classifierPartialUses a model you configure. It marks text safe when that model does not answer.
Inline proxy to Anar GatewayPartialBuilt. No Anar product routes through it yet.
Framework reportsPartialFive frameworks. The score counts activity. It does not test controls.
Names in Arabic text, tenants, tamper evidenceRoadmapOne customer per deployment today.
Anar Guard
Roadmap
Next, in the order a security review asks for it.
Now
  • Keyword and pattern policies
  • GCC personal-data masking
  • Audit trail with export
  • Alerts and signed webhooks
Next Roadmap
  • Safety classifier inside your boundary
  • Arabizi and Gulf dialect detection
  • Names and places in Arabic text
Later Roadmap
  • Human review queue
  • Policy versions and replay
  • Evidence mapped to each control
Anar Guard
Put your rules in front of the model.

A one-month pilot: we deploy Guard in your environment and connect it to one team's AI use.

anarlabs.ai/contact

Anar Guard