Industries · 06

AI & developer tools when the model can call your APIs.

LLM features, MCP servers, agents and developer platforms. We test what happens when untrusted text can reach tools, data and infrastructure, and help you build it safely.

Overview

Once a model can read untrusted content and call tools, every document, email and web page is a possible instruction.

AI features introduce a new class of vulnerability: prompt injection, data leaking through context, agents acting on behalf of the wrong user, and tool servers that trust whatever the model sends. They sit on top of the usual risks in APIs, authentication and cloud.

We research this area and publish what we find, including our MCP server security checklist and our write-up of the Ollama vulnerability known as Bleeding Llama. That research goes straight into how we test and build.

Frameworks we work to

  • OWASP LLM Top 10
  • OWASP ASVS
  • OWASP API Top 10
  • MITRE ATLAS
  • NIST AI RMF

Risks

Where AI features
go wrong.

What our testers look for first in ai & developer tools, from experience rather than a generic checklist.

  1. 01

    Prompt injection

    Instructions hidden in documents, emails or web pages that the model follows.

  2. 02

    Tool and MCP access

    Tool servers with broad permissions, no authentication or no checks on what the model asks for.

  3. 03

    Data leakage

    One user's data, system prompts or secrets surfacing in another user's answer.

  4. 04

    Agent authorisation

    Agents acting with their own permissions instead of the user's.

  5. 05

    Model hosting

    Inference servers and model APIs exposed to the internet without authentication.

  6. 06

    Supply chain

    Models, packages and plugins pulled in without review.

Evidence

What you get asked for.
And what we give you.

The requests that usually trigger a test in this sector, and the evidence that answers them.

Asked forWhat it meansWhat we provide
Enterprise AI reviewsBuyers now ask how AI features handle their data and resist prompt injection.A report covering AI-specific and conventional findings, with fixes and a retest.
OWASP LLM Top 10The common reference for LLM application risks.Findings mapped to the relevant OWASP LLM categories.
Data protectionPersonal data in prompts, logs and training data still falls under GDPR.Testing of where data flows and who can retrieve it.
Responsible disclosureVulnerabilities in third-party models or tools may need reporting to vendors.We coordinate disclosure with you when we find issues in software you depend on.

NoteThis field moves quickly. We test against current public guidance and our own research, and say so when something is unproven.

FAQ

Questions

From ai & developer tools teams. Anything else, call us.

An attack where text the model reads, such as a document, email or web page, contains instructions that override what you intended. It matters most when the model can call tools or see other users' data.

Yes. We test authentication, authorisation, input handling and what each tool can actually do, and we have published an MCP server security checklist.

Yes. We test your application, prompts, tools and data flows. Vulnerabilities in the model provider's own service are their responsibility, but we can coordinate disclosure if we find one.

Yes. Our developers build LLM features and MCP servers with least-privilege tools, user-scoped access and human approval for risky actions.

Tell us about your platform.
We'll scope the right test.

A security engineer or developer reads every enquiry and replies, not a salesperson.

  • Fixed quote after a short scoping call
  • NDA available before you share details
  • Remediation and retest on every security engagement

Domains are fine as plain text (app.example.co.uk).

We only use your details to reply to this enquiry. See our privacy policy.