Services

AI security review for agents and LLM applications

A fixed-scope review of what your agents and LLM applications can reach, what they can do with it, and what you can prove afterwards. The aim is to reduce risk and give you evidence, not to promise that nothing gets through.

What you get

When this fits

  • You are putting agents or LLM applications in front of real data and real tools.
  • Someone, a customer, auditor or board, will ask how you know it is safe.
  • You want a written, prioritised list of what to fix first.

How it works

01

Short call and written scope

A 30-minute call or an email, then a one-page scope with deliverables and timing. Nothing starts until you agree to it.

02

Review

Two to three weeks looking at tools, permissions, logging, blocking and guardrails around your system.

03

Findings

A prioritised report you can act on. Fixing the findings is scoped separately.

Common questions

What does the review cover?
What your agents and LLM applications can reach and do: which tools and actions each agent is allowed, whether unauthorized tool calls and agent actions are logged and blocked, and what guardrails sit around private and commercial AI and RAG architectures.
What do we receive at the end?
Prioritised findings you can act on. Implementing the controls is scoped separately afterwards.
Do you follow this area closely?
Every weekly issue of this newsletter covers AI and security news, including attacks on agents and AI infrastructure. That reading is part of how we know what to look for.
Does it guarantee we are secure?
No, and we say so up front. The review reduces risk and gives you evidence; no review can promise that nothing gets through.

Related

Have a system to build, secure or document?

Tell us what it is and we’ll get back to you within 48 hours.

Get in touch