AI penetration testing, supervised and sovereign
Autonomous pentesting. Sovereign. Under control. RAPTOR attacks your applications as someone who understands your business would, and only reports what it has proved. It works the same in our cloud, on your premises or fully isolated, with no Internet connection. Spanish technology.
RAPTOR is a provisional name.
Designed not to give false positives
It only reports what it proves.
It understands your business
Roles, permissions and business flows: what a scanner does not see.
Offline as well
It works the same in the cloud, on your premises or isolated.
Fixed price
For a web application, with a report and a retest.
Three differences
Automated pentesting that does not stop at a scanner
Three things a scanner does not give you, and that change what your team receives.
What defines RAPTOR
Offline as well.
It works the same in the cloud, on your premises or isolated.
Cloud · your premises · isolated
How it does it: where it runsIt avoids false positives.
It only reports what it proves.
Verification and an identity differential
How it does it: no false positivesIt understands your business.
Roles, permissions and business flows.
From anonymous access to each role you declare, compared with one another
How it does it: your business
How it does it: false positives, your business and where it runs.
Designed not to give you false positives
Scanners, and many pentests, deliver long lists full of false positives, and your team spends days discarding them. RAPTOR is designed for the opposite: it only reports what it proves. And it leaves nothing half-done: every indication is investigated until it is proved or discarded, so what it reports to you is real.
For technical teams: the detail
It is reproduced
RAPTOR does not stop at the indication: it takes it through to the end and exploits the flaw in a controlled way to prove it or discard it.
It is verified
Up to seven types of check decide whether the proof is conclusive. If it is not, it is not reported as a finding.
It is compared across identities
The same resource, requested as anonymous and with each role you declare: if someone sees what they should not, that is demonstrated.
It is documented
Evidence with its fingerprint and CVSS criticality calculated on that evidence, always with the same criteria.
- Evidence with a fingerprint for each finding.
- CVSS 3.1 and 4.0, with deterministic severity and confidence.
- Up to seven types of oracle and an identity differential that adapts to your application's roles (from anonymous access to each profile you declare) decide whether a flaw is demonstrated.
- Integrated NVD vulnerability database, which works offline.
- OWASP WSTG and Top 10, PTES, CWE and MITRE ATT&CK methodologies.
The card is an illustrative example: it shows how a finding reaches your report. It does not come from any client or real engagement.
A customer can read another customer's invoices at /api/facturas/{id}
- Proof
- Reproduced with two roles: user B obtains user A's invoice
- Evidence
- Request and response stored, with their fingerprint
- Criticality
- High (CVSS 4.0), calculated on the evidence
- Impact
- Exposure of billing data for every customer
- Fix
- Check on the server that the invoice belongs to the session user
It understands your business: roles, permissions and flows
The flaws that hurt most are not a known signature: they are a customer who sees someone else's invoices, or a user who reaches administrator functions. Finding them means understanding who can do what, and a scanner does not know that. RAPTOR does prove it: it walks the application's flows with each role, from anonymous access to each role you declare, and compares what each one sees and what each one can do.
Roles
It tests the application as anonymous and with each role you declare.
Permissions
It detects when a user reads or modifies another user's data, or reaches functions their role should not have.
Business flows
It walks the application's flows with each role, not only isolated pages, and compares what each one sees and what each one can do.
For the more particular business rules of your organisation, RAPTOR works alongside our pentesters.
How RAPTOR works
How an engagement works
Supervised autonomous pentesting: RAPTOR works on its own, always within the engagement's controls. Scope is defined with an allow-list and a deny-list, and it is applied on every request, before it is sent.
Phases, controls and a sample run
Allow-list: what may be touched
The authorised hosts, ports and paths. Outside that list, RAPTOR sends nothing.
Deny-list: what is never touched
Express exclusions inside the scope, such as a payment gateway or a function that deletes data.
Windows and rate limit
It works only in the agreed windows and does not exceed the rate you set.
Stop and record
The engagement can be stopped at any time, and every decision is recorded with its reason.
Where it works: in the cloud, on your premises or offline
RAPTOR works the same in all three: same engine, same results and same controls. Working offline is an option for those who need it, not a condition.
Private company
Cloud
Nothing to install: RAPTOR works from our platform, with the same controls for scope, windows, rate and stop.
If you want it in-house
On-Premise
RAPTOR is installed on your premises, on an appliance supplied by Minery Report. To update, it uses only an allow-list of destinations.
Optional, for isolated networks
On-Premise Offline
On your premises and fully isolated, with no Internet connection. How it works
In all of them: masked evidence, deletion per engagement and isolation between clients. And in all of them, RAPTOR improves with each version.
Defence
Defence: sovereign AI to attack and to defend
RAPTOR combines offensive and defensive capabilities with AI of its own design and, when needed, works on isolated networks that handle classified information, where no tool can call home.
Offensive, defensive, offline and sovereignty
Offensive
AI penetration testing of your systems
The same agents that attack a web application, with capabilities on Active Directory and Linux and Windows systems, within the scope you set.
Defensive
Detection ready for your SOC
An agent specialised in detection that generates different indicators of compromise (IOC) from each finding, so your SOC can detect the attack if someone tries to repeat it.
Offline option
On-Premise Offline
No Internet egress, imposed by the appliance's own network: no data leaves. The AI model and the vulnerability database are inside.
Sovereignty
Sovereign AI, of our own design
Designed and developed in Spain by Minery Report, with its origin in the Spanish Air and Space Force. Appliance supplied by Minery Report, with an annual licence and support.
Public administration
Public administration: aligned with the Spanish National Security Framework (ENS), and your data stays in
For bodies that need to test their systems without information leaving the premises, and with evidence that serves in their audits.
More on how we work with the public sector (in Spanish).
ENS, data stays in, and a Spanish manufacturer
Aligned with the ENS. The evidence of each finding helps you demonstrate the technical tests required by the Spanish National Security Framework (ENS, in Spanish).
Your data stays inside. On-Premise, it works on your premises; in On-Premise Offline, also with no Internet egress: no data leaves.
Spanish manufacturer. Annual licence with maintenance, support and continuous updates, and a report with your organisation's branding.
What it audits: web applications first
Its strength is web applications and their APIs: more than 60 web probes and more than 200 integrated and governed tools, spread across more than 12 specialist agents.
See the more than 12 specialist agents
- Reconnaissance
- Web
- Active Directory
- Cloud
- Mobile
- Wireless
- Social engineering
- Exploitation chains
- Proof-of-concept validation
- Detection
- Report
- Language-model red team
- And many more…
What we tell you plainly
- Active Directory and Linux and Windows systems of multiple versions are agent capabilities, with variable maturity and more operator intervention than on the web.
RAPTOR against a vulnerability scanner
Both are automatic. The difference is what you can do with the result.
| Aspect | Conventional scanner | RAPTOR |
|---|---|---|
| What it delivers | A list of alerts | Proved findings, with their evidence |
| False positives | Frequent | Designed not to give them: it only reports what it proves |
| Business | It does not understand roles or flows | It tests roles, permissions and business flows |
| Severity | Estimated | Calculated on the evidence |
| Control | It is launched and forgotten | Allow-list, exclusions, windows and rate |
| Where it works | In the vendor's cloud | Cloud, your premises or isolated, the same in all three |
| Report | The vendor's template | With your company's brand |
| Origin | Foreign vendor | Spanish |
What you get
A report with defensible evidence, ready to show to management, to a client or to an auditor.
A report with your brand
For management and for the technical team, in PDF, DOCX, HTML or Markdown.
Findings with their fix
Each one with the proof, the impact and how to fix it, ordered by criticality.
Decision record
What the platform did and why, so you can review it or show it to an auditor.
Indicators of compromise
Indicators of compromise (IOC) so your SOC can detect and stop the attack if someone tries to repeat it.
For your compliance
It is aligned with the ENS (in Spanish), NIS2 (in Spanish) and ISO 27001 (in Spanish) and it helps you demonstrate the technical tests they require.
And if a finding ends up in court proceedings? That is already work for our digital forensics and expert witness team (in Spanish), with its own procedure and its chain of custody.
Designed by people who compete at the highest level
Hack The Box is the international reference platform for ethical hacking training: machines that are vulnerable on purpose, and a worldwide ranking of those who manage to compromise them. The Minery Report team that designs RAPTOR is in the top 0.55 % of that ranking (October 2026). RAPTOR's exploitation logic does not come from a theoretical laboratory: it is written by operators who compete at the highest level.
Behind it is Minery Report: RAPTOR is designed and developed in Spain, with its origin in the Spanish Air and Space Force.
Three ways to start
We do it for you
A web pentest with RAPTOR, at a fixed price, with a report and a retest. See the RAPTOR pentest (in Spanish)
A licence for your team
Annual licence with maintenance, support and continuous updates: new probes, agents and model improvements reach your deployment without changing how you work. On On-Premise and On-Premise Offline, Minery Report supplies the appliance.
Managed service
Our pentest team operates it for you, within our offensive security (in Spanish).
Frequently asked questions about AI penetration testing
A scanner compares your system with known signatures and returns alerts, many of them false positives. AI penetration testing such as RAPTOR chains tests as an attacker would, understands your application's business (roles, permissions and flows) and only reports what it manages to prove, with its evidence.
It is designed not to give them to you, and it leaves nothing half-done: every indication is investigated until it is proved or discarded. What reaches the report has been reproduced, verified with up to seven types of check and, when roles are involved, compared across identities. What it reports to you is real.
It understands roles, permissions and business flows: it walks the application's flows as anonymous and with each role you declare, and compares what each one sees and what each one can do, to find improper access between users or to functions that are not theirs. For the more particular business rules of your organisation, it works alongside our pentesters.
Yes, if you need it: it is an option, not a condition. RAPTOR works the same, and with the same results, in our cloud, on your premises or fully isolated. In the On-Premise Offline mode the appliance has no Internet egress, imposed by its own network, and no data leaves: the AI model and the vulnerability database are inside.
It depends on how you want it. If you need to test a web application, the RAPTOR web pentest (in Spanish) has a fixed price and we do it ourselves. If you want the platform for your team, it is an annual licence with maintenance, support and continuous updates, quoted according to the mode and the scope.
The run is measured in hours: in a sample engagement it was 2 h 43 min. We set the report delivery time in the proposal.
The scope: an allow-list of hosts, ports and paths that may be touched and a deny-list of what is never touched. A RAPTOR agent orchestrates the exclusions, applies them on every request, before it is sent, and respects the agreed windows and rate limit. Always with written authorisation from the system owner.
It depends on the mode. In On-Premise Offline the appliance has no Internet egress: nothing leaves, by design. In On-Premise everything is processed on your premises and the only external connection is the platform update, limited by allow-list. In Cloud the analysis runs on our platform, with encryption in transit and at rest, isolation between clients, masking of sensitive evidence and verifiable deletion on request.
It helps you demonstrate the technical tests they require: it works with OWASP, PTES, CWE and MITRE ATT&CK and is aligned with ENS, NIS2 and ISO 27001.
The model runs inside the platform and is interchangeable: RAPTOR does not depend on a single AI provider and accepts any language model (LLM).
Want to see it against your own application? Tell us what you want to test and in which mode, and we will prepare the demonstration.
Request a demo