Penetration Testing
Manual, objective-driven testing of web applications, APIs, and networks (internal, external, wireless, and physical), built on hands-on consulting experience since 2018, not a scanner with a report template.
Service 04
AI-assisted penetration testing
Most of a penetration test's calendar time is mechanical (enumeration, correlating scattered findings into an attack path, writing it up), not the exploitation work that actually needs a human. Agentic tooling compresses the mechanical part, so more of the engagement goes to the judgment calls that matter, and findings reach you faster.
Overview
The problem we are solving
This is not a chatbot wrapped around a scanner. AI assistance here means directed, human-supervised agentic tooling used to accelerate specific tasks: mapping attack surface, correlating a misconfiguration here with a credential there into a coherent path, drafting reproduction steps, while every exploit that lands in your report is still run and confirmed by hand.
Infrastructure is chosen based on what your data can tolerate. Most engagements run through an isolated Amazon Bedrock deployment inside our own dedicated AWS environment, never a shared consumer product and never used to train anything. For environments where no inference can leave the building, the same agentic workflow runs against a private model on bare metal we operate in-house, trading some capability for a hard guarantee that nothing leaves client-controlled infrastructure.
InFocus is Daybreak and Anthropic Offensive Trust approved: explicitly recognized to use frontier AI models for authorized offensive security testing, rather than operating in the gray area most vendors advertising "AI pentesting" currently occupy.
Coverage
Where AI assistance is used
Scope is agreed up front; these are the areas an engagement typically covers.
- Reconnaissance & attack-surface mapping
- Enumeration across hosts, services, and identities, correlated and prioritized far faster than manual review alone: the same ground a human would cover, covered sooner.
- Evidence correlation & attack-path building
- Chaining scattered, individually minor findings into the realistic path an attacker would actually take, surfaced as leads for the consultant to pursue by hand.
- Payload & test-case generation
- Candidate inputs and permutations generated for manual review and execution: a starting point for the tester, never something run blind against your environment.
- Report drafting & reproduction steps
- Verified findings turned into clear technical writeups faster, so review time goes to accuracy rather than formatting.
- Exploitation & validation: human-led, always
- Every exploit that reaches your report is performed and confirmed by a person. AI output is a lead until a human verifies it; it is never reported as a finding on its own.
Methodology
How the engagement runs
-
01
Infrastructure selection
Scoping decides whether AI assistance runs through an isolated Amazon Bedrock deployment in a dedicated AWS environment, or a private model on in-house bare metal, based on what your data-handling requirements actually allow.
-
02
Human-directed tasking
The consultant directs every task given to the model. Nothing runs autonomously against your environment without a human reviewing the plan first.
-
03
Verification before anything is reported
Every AI-surfaced lead is manually validated and, where applicable, exploited by hand before it becomes a finding. Nothing from the model reaches your report unverified.
-
04
Standard reporting & debrief
The same deliverables as every InFocus engagement: ranked findings, reproduction steps, a live debrief, and validation instructions your team can run on its own.
Deliverables
What you receive
- Full technical report with ranked findings, evidence, and reproduction steps
- A plain statement of which infrastructure handled AI assistance on your engagement, and what did and did not touch it
- Validation steps for every finding, so your team can confirm and close out each fix on its own
- Live debrief with engineering and leadership
Aligned to
- Anthropic usage policies for authorized security testing
- AWS Bedrock shared-responsibility & data-handling model
- MITRE ATT&CK
Questions
Frequently asked
Does using AI mean a person isn't really doing the testing?
No. Every exploit and every finding in your report is manually verified. AI accelerates the mechanical parts of the engagement (enumeration, correlation, drafting); a person performs and confirms exploitation.
Where does our data go when AI assistance is used?
Either an isolated Amazon Bedrock deployment inside our own dedicated AWS environment (never shared, never used to train anything), or a private model running entirely on bare metal we operate in-house, with nothing leaving client-controlled infrastructure. You choose during scoping.
Are you actually allowed to use AI models for offensive security work?
Yes. InFocus is Daybreak and Anthropic Offensive Trust approved: explicitly recognized to use frontier models for authorized offensive security testing under the vendor's own usage terms, not operating against them.
Does this cost less than a standard engagement?
It mainly buys back time rather than cutting price directly. It is part of what makes the Rapid Assessment's five-day timeline possible, and it shortens the mechanical phases of a standard penetration test as well.
Can this be added to an engagement we're already scoping?
Yes, ask during scoping. It is an acceleration layer on top of our standard penetration testing and rapid assessment work, not a separate product.
Related
Often scoped together
Rapid Assessment
A fixed-price, five-day internal network assessment focused on the quick wins and heavy hitters (the handful of issues that give real attackers their first foothold) at a fraction of what the larger firms charge for the same depth.
Get started
Ready to scope ai-assisted testing?
Scoping conversations are free and usually take twenty minutes. You will speak directly with the consultant who would do the work.
Request an assessment
or reach out directly, (608) 520-0719 ryan.zagrodnik@infocus-security.com