White label red teaming: securing the AI agents your agency delivers

White label red teaming lets an AI agency or a pentest firm deliver an agent test without building the expertise in house. CleanIssue attacks the agent, writes the report in the agency’s format and stays invisible to the end customer. This page describes why agencies need it, how the collaboration runs, how pricing is built and what the agency provides. The technical content of the test is detailed in the pillar on AI agent red teaming.

Why does an agency that delivers AI agents need a red team?

Because its customers now ask about security before they sign or accept delivery. An agency that builds assistants with n8n, Make, Dust or custom code plugs these agents into email, a CRM, a helpdesk. Each connection adds a possible action, and therefore a risk if hostile text reaches the model.

Four pressures keep coming back:

• Customer security questionnaires. IT departments increasingly ask what protects the agent against prompt injection and unapproved actions. A test report answers better than a statement of intent.
• Contractual liability. If a delivered agent sends a wrong email or exposes data, the agency is the first to be questioned. A documented test before delivery records what was checked and when.
• GDPR. Article 32 requires technical and organizational measures appropriate to the risk, and provides for a process to regularly test and evaluate their effectiveness. An agent that handles personal data falls within that frame.
• The EU AI Act. For high-risk systems, Article 15 requires an appropriate level of cybersecurity and resilience against attempts by unauthorized third parties to alter their use, outputs or performance. Not every agent is a high-risk system, and application dates are staggered: check the classification of each project with your counsel.

These texts do not mention prompt injection as such. They create a context in which skipping a test becomes harder to justify.

What is white label red teaming?

It is a test performed by CleanIssue under your name, on an agent you built or are about to deliver. The report is written in the agency’s format and identity, with no mention of CleanIssue. Your customer receives a document from your firm, and you keep the relationship.

In practice, white label covers:

• a report in your template, with your logo and layout;
• a debrief you can run yourself, or in which CleanIssue takes part as technical support if you ask;
• no direct communication from CleanIssue to your end customer, unless you arrange it.

The report is confidential and no data is kept after the engagement. More formal confidentiality terms, such as a signed agreement, are discussed case by case during scoping.

When does the test fit in the delivery cycle?

Before each delivery to a customer. The ideal moment is the end of acceptance testing: the agent is complete, tools are connected, and nothing is in production yet. We test, you fix, then we replay the attacks that had succeeded.

This rhythm gives the agency three benefits:

• Flaws are fixed before the customer knows about them, in a delivery context rather than an incident.
• The test report can be attached to the delivery file.
• The agency builds up practice: fixes found on one project (tool permissions, human approval, filtering) become standards for the next ones.

For an agent already in production, a pilot engagement on a delivered project measures exposure before rolling the practice out.

What does the re-test include?

A re-test replays the attacks that succeeded, after your fixes. It confirms that each flaw is closed and that the fix did not open another. It is offered as an option and recommended, because a fix on an agent often changes behavior elsewhere.

The result is an updated report: closed findings, remaining findings, and what changed. It is the document the agency can show its customer as proof that the issue was handled.

How does pricing work for an agency?

Pricing is by quote, with a per-agent and per-delivery logic. The Agent Red Team Express at €1,900 for one agent in five days remains the entry point for testing a first agent. For recurring work or white label, the Custom Red Team is quoted according to the complexity of each agent.

What moves the quote:

• the number of agents and their complexity (number of tools, sensitivity of actions, memory, several cooperating agents);
• the number of MCP servers consumed or exposed;
• the number of environments to test;
• the white label format, which requires formatting the report in your template;
• the re-test option.

A framework rate is possible: a price per agent, known in advance, that you build into your own quotes. It is discussed during the discovery call. No price is listed beyond the Express, because the grid depends on the fleet of agents you deliver. See also the pricing page.

How does the collaboration run?

In five short steps, with a single person on the CleanIssue side:

1. Discovery call of 30 minutes, free: your agents, your tools, your delivery rhythm and what your customers ask for.
2. Quote and a test authorization letter, signed by the system owner. If you test a customer’s agent, that customer gives the authorization.
3. Scoping: test access, environment, dummy data.
4. Testing: attacks on the agent, its tools, its data sources and its MCP servers, following the method.
5. Report in your format, debrief, then a re-test if you chose it.

Every test requires written authorization: in France, fraudulent access to an automated data processing system falls under articles 323-1 and following of the Code pénal. This rule also applies when you test a customer’s agent.

What does the agency need to provide?

A test environment and the scoping information. The test runs on staging or a dedicated instance, with dummy accounts and data, never on a customer’s production.

What CleanIssue needs:

• the list of the agent’s tools and their rights;
• access to the agent (chat link, API, test email address) and to the workflow, read-only if possible, to understand the logic;
• a description of the data sources (document base, emails, web pages);
• a technical contact reachable during the engagement;
• the signed authorization letter;
• your report template, if you want white label.

Which agencies and firms is this for?

Two profiles. AI agencies that build agents for small and mid-sized companies on tools such as n8n, Make or Dust, and want to sell a tested agent. Pentest firms that receive requests about AI agents and want to subcontract the agent part while keeping the customer relationship.

In both cases, the n8n webhook case study shows the type of finding produced on an automation workflow. For the customer-side stakes of agent security, see AI agent security.