Red teaming is the AI testing service vendors state most often
By AI Testing Vendors · 2026-11-09
Among published entries, red teaming and adversarial testing is the service stated most often, followed by bias audits and AI security penetration testing. Security firms and QA companies both state it.
Services by how often vendors state them
| Testing service | Published entries | Share |
|---|---|---|
| Red teaming and adversarial testing | 56 | 64% |
| Bias and fairness audit | 44 | 50% |
| AI security penetration testing | 42 | 48% |
| LLM output evaluation | 36 | 41% |
| ML model validation | 31 | 35% |
| Production monitoring and drift testing | 31 | 35% |
| Training data and labeling QA | 25 | 28% |
| Performance and latency testing | 23 | 26% |
| Prompt and model regression testing | 17 | 19% |
| Compliance and conformity assessment support | 16 | 18% |
| RAG pipeline validation | 14 | 16% |
| Conversational flow testing | 11 | 13% |
What red teaming covers
red teaming tries to make a model or AI application produce harmful, unsafe or unintended output. Penetration testing of AI applications targets the system around the model: tools, data access and prompt injection that triggers actions.
red teaming: Deliberate attempts to make a system produce harmful or unintended output. prompt injection: Input crafted to override a model's instructions or extract hidden data.Who states the two services
| Company type | Red teaming and adversarial testing | AI security penetration testing | Both |
|---|---|---|---|
| Pure-play QA company | 14 | 11 | 8 |
| IT services firm with a QA practice | 12 | 8 | 6 |
| AI assurance and audit specialist | 2 | 1 | 1 |
| Security firm with AI testing | 16 | 19 | 15 |
| Crowdtesting platform | 2 | 0 | 0 |
| Data and evaluation vendor | 9 | 2 | 2 |
The guide AI red teaming vs AI penetration testing explains which one to scope first.
Reading the counts
A count shows that a vendor's own AI page names the service. It is a statement by the vendor, dated on each entry page, and not a finding about how well the service is delivered.
Counts in this post are recomputed from the directory at every build; the title and summary describe the data on the post date.