Meaningful adversarial testing of an AI deployment requires the corpus, the tool registry, and production-equivalent identity. That access cannot be extended across a border. It is the reason this work has largely not been done in the region — and the reason Averon exists here.
Residency is usually discussed as a property of the system under test. In an assessment it is also a property of the assessment itself — the evidence, the payloads, the extracted records, and the report are all derived from regulated data.
Findings are delivered already mapped to the frameworks your organisation is assessed against, in the language and structure your compliance function uses. The mapping is indicative and confirmed with your team during scoping — the frameworks below are the ones our reporting is structured around.
| Framework | Authority | Applies to | Averon reporting alignment |
|---|---|---|---|
| Essential Cybersecurity Controls (ECC) | NCA — KSA | Government entities and critical national organisations | Findings mapped to ECC domains, principally identity and access management, data protection, penetration testing, logging and monitoring, and third-party and cloud. |
| Cloud Cybersecurity Controls (CCC) | NCA — KSA | Cloud-hosted deployments and their service providers | Applied where inference, retrieval or orchestration runs on cloud infrastructure, including provider-side model services. |
| Critical Systems Cybersecurity Controls (CSCC) | NCA — KSA | Systems designated critical by their operating entity | Applied where the AI deployment influences or acts upon a designated critical system. |
| Cyber Security Framework | SAMA | Banks, insurers and financial institutions under SAMA supervision | Findings additionally expressed against the framework's identity and access, data protection, and third-party domains for supervisory reporting. |
| Personal Data Protection Law (PDPL) | SDAIA — KSA | Any deployment processing personal data | Exposure findings assessed for personal-data content, lawful basis, minimisation and cross-border transfer implications. |
| AI Ethics Principles | SDAIA — KSA | AI systems deployed by Saudi entities | Referenced where findings bear on transparency, accountability, or human oversight of the deployed system. |
| Regional equivalents | UAE · Qatar · Bahrain | GCC deployments outside the Kingdom | Reporting structured against the applicable national authority's controls for engagements elsewhere in the Gulf. |
Averon is an independent assessor. We do not certify, accredit, or licence, and an Averon engagement is not a substitute for any mandated audit or authorisation. What we provide is technical evidence of control effectiveness, structured so the accountable party can rely on it in their own submission.
Safety training, content filtering and guardrail evaluation are overwhelmingly developed and benchmarked in English. The same control frequently behaves differently when the input is Arabic — and an attacker in this region has no reason to write in English.
Averon tests in both languages as a matter of standard scope, not as an add-on.
Executive-facing artefacts — the launch position, the entitlement differential and the control mapping — are issued in Arabic and English, both authored rather than machine-translated, and both reviewed by the engagement lead.
تُصدر التقارير التنفيذية باللغتين العربية والإنجليزية، وتُحفظ جميع بيانات الاختبار داخل المملكة.
الاختبار السيبراني الهجومي لأنظمة الذكاء الاصطناعي — قبل التشغيل وبعده.
Where a deployment applies input filtering, that filter's effectiveness is not necessarily equivalent across languages. The asymmetry is easy to miss, because the evaluation that produced confidence in the control was usually conducted in English.
This is not a commercial argument. It is a structural one about what the work requires.
The engagement requires the real corpus and production-equivalent identity. A vendor that cannot be granted that access can only test the interface — which is the one part of the system that was already tested.
Evidence derived from regulated data is itself regulated data. An engagement whose artefacts leave the jurisdiction creates the exposure it was commissioned to find.
A report that must be reworked before it can be submitted has not finished. Mapping to the applicable national controls, in Arabic, is part of the deliverable — not an exercise left to the client.
The independent, in-region security layer for organisations deploying AI where sovereignty, data residency and regulatory alignment are not negotiable terms.
Averon — positioningScoping sessions are held in Riyadh or on your premises, in Arabic or English.