How we use AI in penetration testing

Researchers lead. AI widens the coverage.

Attackers already use AI. We account for that when assessing your defenses and use our in-house agentic pentest harness to help our researchers examine more of your environment.

What our harness does

On web and API pentests, it supports reconnaissance, threat modeling, pentest planning, security testing within scope, and drafting findings. Our researchers verify every finding we report.

Authorized scope

We define what the harness may test.

Permitted risk

We set the permitted level of testing risk.

Run limits

We cap the number of runs before testing starts.

Vetted access to cyber-capable models through Anthropic’s Cyber Verification Program and OpenAI’s Daybreak Trusted Access for Cyber. Blaze is a signatory to the CREST AI Charter.

Data sovereignty

We choose the AI hosting setup. You can steer it.

We normally use cloud-hosted models. Tell us during scoping if you have a hosting preference, data residency requirements, or want a pentest without AI.

Commercial AI service

An external provider runs the model under its service terms.

Rented inference

Blaze runs the model on rented data center infrastructure.

Local hardware

Blaze runs the model on hardware we operate directly.

Depending on the setup, we also use OpenRouter, RunPod, and Fireworks AI for model routing and inference.

Where your data can be handled

Pentest infrastructure, model processing, and evidence storage can be in different locations.

Frequently asked questions

Yes. Tell us when we scope your engagement.

It depends on the hosting setup. If you need the provider list for a vendor review, ask during scoping.

Redaction before a request changes what the model receives. Redaction when saving logs changes what Blaze keeps in those logs.

AI use and data handling for vendor reviews