AI red teaming companies

AI red teaming companies sell three different things under one name. The table lists providers whose own pages state AI application or LLM testing, in A to Z order with no ranking. Each row shows the facts we could source, and "Not published" where a provider says nothing. Facts were checked on 9 November 2026.

What does this list include, and how does a provider qualify?#

A provider qualifies when its own pages state testing of AI applications or LLM systems. The dataset records that as "yes" in the AI apps column or as an AI and LLM service. We do not score quality, so qualifying means "states the service". The list is not a ranking, and no provider pays for a position.

NIST AI 600-1 defines red teaming as "A structured testing exercise used to probe an AI system to find flaws and vulnerabilities such as inaccurate, harmful, or discriminatory outputs, often in a controlled environment and in collaboration with system developers." Providers use the word for work that differs a lot.

OWASP's GenAI Red Teaming Guide names four areas: model evaluation, implementation testing, infrastructure assessment and runtime behavior analysis.

Who does this list help?#

It helps teams that ship an LLM feature, a copilot, a retrieval system or an agent, and need an outside test. It also helps security teams that answer customer questions about AI risk. Teams that only call a hosted model through a plain API may need an application test and not a model test.

For the narrative comparison and how to choose, read AI red teaming providers. For plain AI pentest providers, see AI penetration testing companies.

What should you ask AI red teaming providers?#

Ask for these answers in writing before you sign.

Our guide questions to ask a pentest provider has the longer list.

What drives the price?#

Price follows the number of models, tools, data sources and user roles in scope, the depth of testing and the retest terms. Many providers quote after scoping. The "Published price" column shows a figure only where a provider lists one, so a blank price is a missing fact. See penetration testing cost for the ranges we sourced.

How do you read the table?#

Open a provider row for its sources and the date we checked them. The AI apps column shows "yes" only where the provider names that scope. The Sample report column says whether a report is public, on request or not published. Read how we compare for the rules. For the service-level view, see AI and LLM penetration testing.

What are the limits of this list?#

The list shows what providers state. It does not show how good their tests are. OWASP's LLM01:2025 entry says that "it is unclear if there are fool-proof methods of prevention for prompt injection", so no test result proves a system is safe. A provider that claims otherwise has gone past the evidence.

Many AI-focused vendors sell software and not a scoped test. Read the type column before you compare quotes. The EU AI Act uses the term "adversarial testing" in Article 55(1)(a), and we did not find a rule that names red teaming for all AI systems.