How it works
From request to signed letter
Six steps. You do two of them, and neither takes longer than a coffee break.
Tell us what to test
The request form asks what the system is, which domains it runs on, who hosts it and whether another company builds or runs it for you. None of it is a trick question: if you don't know an answer, write that and we'll work it out together.
You also sign a short written authorization. It says you are allowed to ask for this test, that we may only test what is listed, only from the internet, and only during the window we agree. The exact wording is on the form, and the version you agreed to is stored with your request.
Takes
5 minutes
You do
Fill in one form. About five minutes.
Show us it's yours
We show you a TXT record to add to each domain: the name is _provensecure and the value is a random token that belongs to your request only. Whoever manages your DNS can add it in a couple of minutes.
Press “Check now” on your request page and we look the record up at the public DNS resolvers. One record on your main domain also covers its subdomains, so example.com proves app.example.com too.
Nothing is tested before every domain checks out. This is the step that makes it impossible to point us at someone else's systems.
Takes
Minutes to an hour, depending on your DNS provider
You do
Add one DNS TXT record per domain, then press “Check now”.
We agree scope and dates
We write back with the scope as we understood it, anything we would like to exclude, the rate limits we will hold to, and the testing window. Testing starts only once you have confirmed that.
If your system is run by a third party whose terms require their permission, this is where we check it. If you can give us test logins, we use them to test what a signed-in user can reach, which is where the serious problems usually are.
Takes
1–2 working days
You do
Reply to one email.
We try to break in
We map everything of yours that is reachable from the internet, then work through a structured methodology (OWASP WSTG for web and APIs, the OWASP LLM Top 10 for AI features) by hand, with tooling where tooling helps.
A problem only counts once we have demonstrated it: a command that shows the weakness, plus a control run without the attack that behaves differently. Both are saved as evidence. Anything we merely suspect stays a hypothesis and never reaches your bill.
If we find something critical, you hear about it the same day by phone or email — you do not wait for the report.
Takes
A few working days for a typical site or app
You get the full report, free
The report opens with a plain-language summary your board can read, then gives engineers the reproduction steps, the evidence and a concrete fix for each finding. Severity combines technical impact with what the problem means for your business.
Medium, low and informational findings are in there too, at no charge. You see everything we found before you are asked to pay anything.
You can see exactly what you will get: the sample report on this site is a complete example for a fictional company.
Takes
1–2 working days after testing ends
Fix it, then prove it
A retest re-runs the exact command that demonstrated each problem, plus the obvious ways around the fix. You get an updated report showing what is now closed.
If the retest is clean and the planned scope was completed, we issue an attestation letter with a code and a QR link to a public verification page. Anyone you show it to can check it is genuine, and we can withdraw it if it stops being true.
The letter states what was tested and what was proven on the date it was issued. It never claims your systems are secure in the absolute, because no honest letter can.
Takes
Whenever you're ready
You do
Fix the findings, then ask for the retest.
What we actually look at
We follow a published methodology rather than whatever a scanner happens to flag: the OWASP Web Security Testing Guide for web and APIs, and the OWASP LLM Top 10 for AI features.
Websites and web apps
Access control between accounts, authentication and sessions, injection, file handling, business logic you can abuse, and the configuration around all of it.
APIs
Every endpoint we can reach, with and without credentials: object-level authorization, mass assignment, rate limits, token handling and data exposed beyond what the client shows.
AI assistants and agents
Prompt injection, tool and function abuse, data exfiltration through the model, and exposed MCP or agent interfaces — mapped to the OWASP LLM Top 10.
The cloud around them
Publicly exposed storage, leaked credentials in responses or builds, forgotten subdomains and services nobody remembers switching on.
Proven, not guessed
A finding only reaches your report once we have demonstrated it. That is also what makes the bill honest: we can only charge for problems we proved.
- 1We run the attack. A specific command or request that shows the weakness, saved exactly as it ran.
- 2We run the control. The same call without the attack condition. It has to behave differently, or the finding does not stand.
- 3We keep the evidence. Request, response, the difference between them and any files, stored with the engagement.
- 4You can re-run it. Every finding ships with the steps to reproduce it yourself, which is also how the retest works.
What we never do
- Test anything you haven't proved you own. Every domain needs a DNS record only its owner can add, plus written authorization naming it.
- Touch internal networks or office systems. We test only what is reachable from the public internet. No VPNs, no laptops, no office Wi-Fi.
- Phish or call your staff. No social engineering of any kind, unless you commission it separately and in writing.
- Run attacks meant to take your service down. No denial-of-service, no load testing. We hold to an agreed request rate and stop if your system starts to struggle.
- Change, delete or keep your real data. Read-only by default. Anything we do create during a test is listed and removed afterwards.
- Share your findings with anyone you didn't name. The report goes to the contacts in your request. Evidence is deleted 90 days after the final report.
Ready when you are.
Start with the form. Nothing is tested until you have proved the domains are yours and agreed the dates in writing.