AI security testing in Abu Dhabi
- Data regime
- Federal UAE data protection law, with free zones such as DIFC and ADGM operating their own regimes. Which applies depends on where the entity is registered.
- Working hours
- GST, three hours ahead of CET. A full morning of overlap, with the UAE working week running Monday to Friday.
- Languages
- Arabic and English, frequently in the same system.
- Delivery
- Remote testing against your staging or production endpoint, with scoping and findings sessions scheduled in the Abu Dhabi working day.
- Frameworks
- OWASP Top 10 for LLM Applications, MITRE ATLAS, NIST AI RMF, EU AI Act Article 15.
- Tools
- Garak, PyRIT, promptfoo, Giskard, Burp Suite and custom Python harnesses.
Authorised testing only
Every assessment runs under a written scope and authorisation from the owner of the system, agreed before any test is run.
Why AI security matters in Abu Dhabi
ADGM data protection
Entities in Abu Dhabi Global Market fall under the ADGM Data Protection Regulations, modelled on GDPR. An AI leak of personal data is assessed under those rules.
Government-grade assurance
Government and government-linked entities work to demanding security standards, and AI assistants connected to their data are expected to meet them.
Open-weight and sovereign models
Abu Dhabi organisations are more likely to run open-weight or locally developed models on their own infrastructure, which adds model file security and self-hosted guardrails to the scope.
Arabic-first systems
Many assistants here are Arabic-first. Attacks are written in Arabic from the start rather than translated from English.
What gets tested
The six attack classes behind most serious findings. The full list, and the tooling used for each, is on the main AI security page.
Direct prompt injection and jailbreaks
Role-play, instruction override, payload splitting, encoding tricks (Base64, leetspeak, invisible Unicode) and multi-turn escalation, to see whether the model can be argued out of its instructions and guardrails.
Indirect prompt injection
Instructions planted in the content your system reads rather than in the chat box: an uploaded PDF, an inbound email, a web page an agent browses, a product review, a CRM note. This is the attack most production systems are least prepared for.
System prompt and configuration leakage
Extracting the hidden instructions, internal URLs, API structure, business rules and occasionally the credentials that developers put in a system prompt on the assumption nobody would see it.
Data exfiltration through output
Markdown images, auto-unfurled links and tool calls that quietly send conversation data or retrieved documents to an attacker-controlled server once a malicious instruction lands.
Excessive agency and tool abuse
Agents persuaded to send emails, issue refunds, change records or call internal APIs outside their intended purpose. Tested against the real tool permissions, including MCP servers and poisoned tool descriptions.
RAG and vector store weaknesses
Cross-tenant document leakage, retrieval that ignores the user's access rights, poisoned documents that steer answers, and embeddings that reveal more than the source permissions allow.
How the assessment runs
1. Scope and threat model
Map what the AI system can read, what it can do, who talks to it and what would hurt most if it went wrong. Written authorisation and rules of engagement are agreed before any testing.
2. Automated scanning
Garak, PyRIT and promptfoo run thousands of known attack patterns against the live or staging endpoint to establish a baseline quickly and cheaply.
3. Manual adversarial testing
The part that finds the serious issues: multi-turn manipulation, indirect injection through your real document and email flows, and chained attacks that scanners cannot plan.
4. Agent and integration testing
Every tool, API and permission the model can reach is tested for abuse, including privilege boundaries between users and tenants.
5. Report and fixes
Each finding comes with a reproduction, a severity, the OWASP LLM and MITRE ATLAS mapping, and a concrete fix — architecture first, filters second.
6. Retest and regression suite
Fixes are retested, and the successful attacks become a promptfoo suite in your pipeline so they cannot quietly come back with the next model upgrade.
Areas served
AI security testing is available across Abu Dhabi, including Al Reem Island, Masdar City, Yas Island, ADGM, Khalifa City, Al Maryah Island, Mussafah. There is no local office — testing is delivered online, which is how AI endpoints are attacked in practice anyway.
AI risk in Abu Dhabi's key sectors
Where AI is being deployed fastest here, and the risk tested first in each sector.
Assistants over accounts, claims and policies are tested for data leakage, manipulation into actions and unsafe advice, with results documented for DORA, the AI Act and your regulator.
RAG assistants over client files and matters must respect confidentiality between clients and between teams. Retrieval access control and exfiltration through rendered output are the priority tests.
Booking and concierge bots are public and connected to reservations. They are tested for price manipulation, leakage of other guests' details and unauthorised booking changes.
Property assistants hold lead data, owner details and pricing rules. Prompt injection that exposes other clients or commits to prices is tested before the bot goes public.
Assistants over shipment, customer and pricing data must not leak one customer's data to another or be steered into changing records. Documents and emails the system reads are tested as injection vectors.
Frequently asked questions
Can you test self-hosted models?
Yes. Self-hosted deployments add checks on model files, inference endpoints and guardrail configuration on top of the usual prompt injection and leakage testing.
Is the work remote?
Yes, from Europe, with most of your working day overlapping.