HOME / SERVICES / AI SECURITY
AI EXPOSURE REVIEW
That is not a criticism. Agents arrive through assistant rollouts, developer tooling, vendor integrations and people solving real problems with whatever works. The AI Exposure Review answers one question in days: what are we actually running, and what can it do.
MOST ORGANISATIONS CANNOT NAME EVERY AGENT RUNNING INSIDE THEM.
DEFINITION
WHAT THE REVIEW COVERS
- Shadow AI discovery, including the endpoint layer. Network proxies and cloud access brokers monitor the wrong layer for AI running locally on a device, which is where a large share of unapproved use sits.
- Agent and application inventory. An AI bill of materials across the major agent platforms and self-built agents, with an owner recorded against each one.
- Security posture baseline. Model selection, system prompts, tool access, guardrail coverage, over-permissioned tools and untested guardrails.
- Non-human identity exposure. How agents authenticate, whether credentials are shared, which agents can create or instruct other agents, and whether an action can be attributed to a person.
- One live adversarial test. A sample attack against one production agent, run by an operator, so you see what this actually looks like rather than reading about it.
- King V accountability mapping and a board one-pager. Written so a director can act on it without a translator.
WHY THE LIVE TEST MATTERS
Everything above can be produced by a competent assessor with a questionnaire and access. The sample attack is what changes the conversation, because it converts a list of theoretical exposures into one demonstrated fact about your own environment.
THIS IS A PAID ENGAGEMENT
- The Review is a fixed-fee, rand-priced diagnostic delivered in days. It is scoped, it is booked, and it produces a written verdict.
The fee is credited against the engagement that follows
Whether that is an Agentic Red Team, runtime protection or a programme of work. That is a commitment mechanism, not a discount, and it exists so the decision to look does not compete for budget with the decision to act.
FREQUENTLY ASKED QUESTIONS
FAQ
How long does the Review take?
Days. The exact number depends on the size of the estate and how quickly access is granted, and we confirm it in writing before we start.
What access do you need?
Read access to the environments where agents run, identity and access logs, and a named technical contact. The live adversarial test requires separate written authorisation naming the target agent and the testing window.
What happens in the thirty minute consultation?
We talk about your AI roadmap and your obligations, and we tell you where we think your exposure sits. It is a scoping and orientation conversation, not an assessment, and no testing takes place. If a Review is warranted we scope it from there.
What if the Review finds nothing significant?
Then you have a dated inventory, an attribution picture and a board one-pager saying so, which is exactly what King V periodic assurance and Joint Standard 2 control testing evidence require. A clean result is a deliverable.
TAKE ACTION
TEST IT BEFORE SOMEBODY ELSE DOES
Thirty minutes on your agent estate and what this risk looks like in your environment.