AI pentesting all year round, with a pentester behind every finding.
Offensive agents running on Carpent, our platform, attack your applications and perimeter continuously. They test authentication, permissions and business logic the way an outsider would. Before a finding reaches your team, a Dasenda pentester reproduces it and attaches the working exploit.
How it works
Four phases that repeat for as long as the service runs.
- 01
Reconnaissance
The agents map what you expose to the internet. Subdomains, published services, routes, forms and APIs, always inside the scope we sign with you.
- 02
Attack
They test what a scanner misses. Authentication bypasses, access to other users' data, injection, business logic flaws and exposed cloud services.
- 03
Validation
A pentester reproduces every finding. If it cannot be exploited, it does not reach you as a vulnerability. Critical issues are reported the same day.
- 04
Fix verification
Once you fix it, the agents attack that point again and confirm in writing that it is closed.
Control
An AI that attacks your systems has to account for every step.
- 01
Scope checked twice
Every action is validated against the agreed scope on the platform and again on the machine that runs it. If the target is out of scope, the action is blocked and logged.
- 02
Exploitation only with your approval
You can require approval before any exploitation phase starts. Without it, the agents stop at detection.
- 03
Every command on record
Everything the agents run goes into a log that cannot be edited or deleted. You know what was done, when, and against which system.
Manual or AI
A manual pentest goes deep once a year. AI pentesting covers the rest of the year.
Most companies run one pentest a year and ship changes every week. Whatever goes live after the report stays untested until the next one. That is why we combine them.
| Manual pentest | AI pentesting | |
|---|---|---|
| Frequency | One or two windows a year | Continuous, for as long as the service runs |
| Strongest at | Long attack chains, complex business logic and social engineering | Authentication, access control, injection and internet exposure |
| New changes | Tested in the next window | Tested when they appear on your exposed surface |
| Verifying a fix | Agreed retest or a new window | Automatic once you fix it |
| Who signs off a finding | The pentester | The pentester, on what the agents find |
Common questions
What people ask about AI pentesting.
- What is AI pentesting?
- It is a penetration test in which AI agents run reconnaissance, attack and verification continuously, with the same tools a pentester uses. At Dasenda a person reproduces every finding before it is delivered, so you receive exploitable vulnerabilities rather than a list of possible issues.
- Does it replace a manual pentest?
- No. The agents cover the exposed surface and the most common flaws well, and they do it all year. Long attack chains, very specific business logic and social engineering still need a pentester with time. Most clients combine one manual pentest a year with the continuous service.
- Is it safe to let an AI attack my systems?
- Scope is signed before we start and checked twice on every action. You can require your approval before any exploitation, and every command goes into a log that cannot be modified. Tests that could affect availability are agreed with you and run in a maintenance window.
- Does it count as evidence for NIS2 or ISO 27001?
- Yes. Both require vulnerability management and periodic testing that controls work. The history of tests, findings and fixes is the evidence an auditor, or a client sending you a security questionnaire, asks for.
- How does it relate to Carpent?
- Carpent is our agentic security platform and AI pentesting is one of its modules. In this service we operate it for you. If you have your own security team and would rather use the platform directly, it can also be licensed.