AI pentesting tools vs. a read-only vulnerability scanner
AI pentesting tools let an AI run real attacks on whatever address you give them. Scanr answers the same question, what could an attacker reach?, by only reading, and only on sites and networks whose owner proved they own them. The limits are written in code the AI can’t change.
The three kinds of tool
| AI pentesting tools | Classic scanners | Scanr | |
|---|---|---|---|
| What it may scan | Any address you type in | Any address you type in | Only sites and networks whose owner proved ownership |
| What it does | Runs real exploits; can create users and submit forms | Sends thousands of probes, some of them attack payloads | Reads, plus forms and test sign-ins the owner allows, at most 600 requests a scan |
| Where the limits live | In instructions given to the AI | In settings the user can change | In code, with no setting to turn them off |
| Signs in | Often, and may try passwords | Optional | Only with test accounts the owner adds, once each per scan |
| Role of AI | Chooses and runs the attacks | Usually none | Reviews the results; never touches your site |
| The report | A long technical write-up | A long list, often with false alarms | Where it is, the proof, the fix, and a prompt for your coding agent |
| Record of what was sent | Varies | Varies | Every request, and every refused one |
1. It only works on what you’ve proved is yours
Most scanners take any address. That’s how a tool meant for defense ends up pointed at someone else’s site. Scanr asks for a DNS record and a snippet on the site, the same idea as verifying with Google Search Console, and checks both again before every scan. For networks, an owner or manager approves in writing and the network itself is verified. More on How it works.
2. The rules are code, not a request to an AI
An AI told to “be careful” can be talked out of it, by a user or by text planted on a page it reads. Scanr doesn’t rely on that. Every request goes through one gate in code: read requests only, verified sites only, no cookies, no sign-ins, no links that look like actions, a fixed speed and a fixed total. The AI can’t send a request at all. The full list is on How the scanner behaves.
3. It’s built for live sites
You can run Scanr on the site your customers are using right now. It never sends a form you haven’t allowed, never tries a password, and signs in only with test accounts you add. Some sites do change things on an ordinary link, like one-click cancel or unsubscribe links, so the scanner refuses links that look like that, and links that carry tokens.
4. The report is a to-do list
Each finding says where it is, shows the proof, explains what someone could do with it, and gives the fix in steps. If you build with an AI coding tool, one button copies instructions you can paste straight into it. You tick findings off as they’re fixed and scan again to confirm.
5. You can check its work
Every scan keeps a log of each request it sent and each one it refused to send. If a finding can’t be proved or ruled out, the report says so and keeps it in view.
When you need something else
With two test accounts, Scanr checks whether one customer can open another’s private page. If you need a full penetration test, a PCI scan or a compliance audit, that’s a penetration tester or an approved scanning vendor, not Scanr.
Questions
Is a read-only scan as thorough as AI pentesting?
No, and it isn’t meant to be. A read-only scan finds what anyone on the internet can reach without logging in: exposed files, leaked keys, open admin pages and risky settings. An AI pentest goes further by running real attacks, which also means it can change or break things. Many businesses want the first before they are ready for the second.
Can the AI be talked into attacking my site?
The AI has no way to send a request. Only the scanner’s code talks to your site, and that code allows read requests to verified sites and nothing else. Text on a scanned page can’t change that.
Who is Scanr for?
Business owners, developers and the IT people who look after them: anyone who wants to know what an outsider can reach on a site or network they own, without hiring a penetration tester or running attack tools on a live system.