Your agents, data, and control boundaries remain the starting point.
Averon scopes product integration and assurance work around where your agents and MCP servers run, where data may exist, and who is permitted to access technical evidence.
Your data rules shape the engagement.
AI security testing touches prompts, completions, retrieved records, tool outputs, and identity context. Those artefacts are sensitive by default. Their location, access, retention, and destruction are agreed before testing begins.
- 01 Inside your environment Testing can run on client-provided infrastructure, with your identity controls, logging, encryption, and network policy remaining authoritative.
- 02 Dedicated isolated tenancy Where external infrastructure is appropriate, the engagement uses a dedicated tenancy, client-approved region, encrypted storage, and an explicit egress policy.
- 03 Named-person access Access is limited to the named engagement team. Every grant has a purpose, an owner, and an expiry. Access is revoked at closure.
- 04 Evidence lifecycle Raw evidence is minimised, distributed only to named recipients, retained for an agreed period, and verifiably destroyed when that period ends.
Evidence translated into the framework you use.
The technical finding stays the same; the control context changes by organisation and market. Averon maps evidence to the standards, regulations, and internal control library relevant to the deployment—not to a fixed home-market template.
| Framework | Reach | How it is used |
|---|---|---|
| NIST AI Risk Management Framework | Global | Risk identification, measurement, governance, and ongoing control effectiveness. |
| ISO/IEC 42001 | Global | AI management-system controls, accountability, lifecycle management, and assurance evidence. |
| ISO/IEC 27001 | Global | Access control, supplier security, logging, secure development, and information handling. |
| OWASP LLM & Agentic guidance | Technical | Finding classification and engineering remediation across model and agent failure modes. |
| SOC 2 | Assurance | Logical access, change management, monitoring, confidentiality, and system operations evidence. |
| EU AI Act & GDPR | European Union | Risk-management evidence, data governance, transparency, security, privacy, and impact context. |
| Sector and national overlays | Client specific | Financial services, healthcare, government, critical infrastructure, and local privacy obligations. |
Framework mapping does not manufacture compliance. Every mapped control points back to a reproduced technical observation, its identity context, the resulting exposure, and the architectural remedy.
Language is part of the attack surface.
A system that works in one language can fail differently in another. Tokenisation, retrieval quality, safety tuning, tool interpretation, and instruction hierarchy all change across languages and scripts.
Averon scopes language testing to the deployment's real users, data, and markets. It is treated as a security dimension, not a translation exercise.
Test the system people actually use
- Primary operating languages and relevant regional variants.
- Mixed-script, transliterated, and code-switched payloads.
- Retrieval and entitlement behaviour across multilingual corpora.
- Guardrail differential: the same attack across languages, with the delta reported.
- Tool arguments, structured outputs, and approval flows in localised interfaces.
Security can travel without making your data travel.
International delivery is not a promise to centralise everything. It is the ability to bring a consistent assessment method into different legal, technical, and operational environments while preserving the client's boundaries.
One methodology
The same finding classes, evidence standard, severity logic, and independence commitments apply across every engagement.
Different control contexts
Delivery topology, standards mapping, language coverage, and evidence handling adapt to each organisation.
Named authorship
The people who perform the work sign the findings and remain available to defend the evidence with engineering, audit, or regulators.
Design the engagement around your boundaries.
Start with the architecture, data classification, markets, and operating model. We will define a delivery topology that fits them.
contact@averon.systems