Service 04

AI risk and governance

Governance, threat assessment and policy for AI, including generative AI and large language model (LLM) applications.

Overview

Teams are adopting AI faster than policies can keep up. Copilots, chatbots and AI agents introduce new threats such as prompt injection, sensitive data leakage and excessive agency that traditional controls were never designed for.

We help you inventory where AI is used, rate each use case by risk, assess LLM applications against current threat guidance, and put proportionate governance in place, so the business can move quickly without creating regulatory or reputational exposure.

Why now? Copilots and AI agents can read your email, documents and systems. Without guardrails, a single crafted message can make an assistant leak data or take actions it shouldn't. Governance doesn't have to slow adoption, it makes it safe to say yes.

What we assess

  • AI inventory and risk tiering. Where AI is used, by whom, with what data.
  • GenAI and LLM threat assessment. Prompt injection, data leakage, insecure output and agent risks.
  • Governance and policy. AI acceptable use, roles, approvals and human oversight.
  • AI vendor risk. Due diligence on model providers and AI-enabled SaaS.

What you receive

  • AI risk assessmentUse-case risk ratings with treatments.
  • LLM threat modelMapped to the OWASP Top 10 for LLM Applications.
  • AI governance frameworkPolicy, standards and an operating model aligned to ISO/IEC 42001.
  • Regulatory positionEU AI Act classification and obligations, where relevant.

How the engagement runs

  1. Scope and discoveryAgree objectives, systems in scope and stakeholders. Collect documentation.
  2. AssessmentWorkshops, configuration and design review, threat modelling and control testing.
  3. Risk rating and roadmapRate each risk by business impact and build a prioritized treatment plan.
  4. Briefing and handoverExecutive briefing, detailed report and a working session with your teams.

Questions

We only use Microsoft Copilot or ChatGPT. Do we need this?

Even off-the-shelf AI raises questions about what data employees share, who can see the output and how it is retained. A light-touch assessment and policy usually takes only a few weeks and closes the most common gaps.

Can you help us prepare for ISO/IEC 42001?

Yes. We run gap assessments against ISO/IEC 42001 and help design the AI management system needed for certification.

Ready to know where you really stand?

Book a free 30-minute consultation. We'll talk through your environment, your obligations and where a risk assessment would make the biggest difference.