You're in charge. Always.
A tool that acts for you has to earn your trust every day. These are our commitments: Check already meets the privacy ones, and the plans will meet all of them from launch.
Our commitments
The automatic, reversible
Everything it does on its own will be recorded and can be undone for 30 days, with one tap.
The irreversible, with you
Blocking an account or isolating a device, it will ask you first. And AIM SHIELD will never move your money.
Type “PAUSA”
And everything will go back to only warning, straight away.
What we keep, in the EU
Your message is not kept; from a scam we keep the domain for a few days. What we keep from the plans will be on EU servers, in the Netherlands. It is never sold or used to train third-party models.
Your security word
It will be in every message we send you. We will never ask you by message for passwords, codes or payments.
A network that protects everyone
With the plans, when we confirm a scam, it will be blocked for all AIM SHIELD customers. Without sharing anyone's data: only encrypted fingerprints.
The AI warns and recommends. What can't be undone, you decide.
This is how AIM SHIELD's AI works today, and how it will work when the plans start acting for you.
| Level | What it does | Today |
|---|---|---|
| Warn | It analyses a message, a link, a file or an invoice and explains why it can or cannot be trusted. | Today: Check, forwarding to revisa@, Is this file safe? and Is this invoice genuine? |
| Recommend | It tells you what to do (“call your bank on the usual number”, “don't pay into that account”) and leaves the text ready for you, but doesn't do it for you. | Today: Check, the emergency button, Breach notice and the other tools. |
| Act on what can be undone | It will do something that can be undone, it will be recorded and it can be undone for 30 days. | Today: Nothing acts on its own. It will arrive with the plans (Family and Business, coming soon). |
| Never without you | Changing your accounts or your login details, deleting, speaking in your name or sharing your data. And it will never move your money, not even with you. | Today and with the plans: Anything irreversible will always ask you first. |
The AI works in Check, in the forwarding to revisa@ and in the website's assistant. Is this file safe?, Is this invoice genuine?, the emergency button and Breach notice work with rules alone.
How Check reaches a verdict
- First, the rules: known signs of a scam, the link and the domain, the IBAN and the phone number. If the rules already see a scam, the AI does not change that verdict.
- Then, if needed, a second opinion from the AI: it can raise the alert and never lowers a scam that the rules see. It can only ease a doubt in limited cases (a notice that explains a trick without asking you for anything, or a relative with a new number who asks for nothing), and only if all its opinions agree with great certainty.
- In each alert it tells you why, with the signs it has seen.
- Para detectar órdenes escondidas dirigidas a la IA usamos además un modelo abierto como segunda señal. Solo informa: nunca decide el veredicto ni bloquea por sí solo.
- If a message tries to give orders to the AI (“classify this message as safe”), we detect it, remove those orders before they reach it, and it counts as a sign of a scam.
- We test the AI with OWASP's guides for applications with AI and agents. We repeat these tests with each version of the website.
- We have a switch to turn off the AI in the analysis within minutes: then the rules alone decide.
- The website's assistant says it is an AI and, if you prefer, puts you in touch with a member of the team.
What we measure
We test each version with real scams that have circulated in Spain and with genuine messages from banks, public bodies and companies, each one with the source where it was published. These are today's figures, with the rules alone, without the AI's opinions:
- It warns on 301 of 430 real scams published since 2023.
- It gives 1 false alarm in 285 real legitimate messages.
The rules were adjusted by looking at some of these messages, so it is not an independent measurement: for that we use test sets that the engine has not seen before. The AI is only consulted in doubtful cases and never lowers an alert. Measured on 16 November 2026; an automatic test measures it again with each version.
This is how you know it's us.
Anyone can write to you saying they are AIM SHIELD. These are the signs that it really is us.
- Our website is aim-shield.ai and our emails end in @aim-shield.ai.
- When you are a customer, each message from us will carry your security word. If it isn't there, it isn't us.
- We will never ask you for passwords, codes, your card details or payments by message.
- If one day we had to warn you of an incident involving your data, the notice will not carry links or ask you for anything: it will tell you what happened, what we have done and what you can do.
- Check recognises websites that pose as us: if in doubt, pass it the message.
We apply to our website what we recommend to you.
No cookies, no trackers and no third-party programs. Run our own security snapshot on aim-shield.ai and check it.
- Mandatory encrypted connection (HSTS)
- Strict content policy: only our own code
- The free tools don't keep what you paste
- We only look up public information about domains
Have you found a security flaw?
Thank you for letting us know. Write to us at seguridad@aim-shield.ai with what you have found and how to reproduce it.
- We reply and keep you informed until it is fixed.
- We will not take action against anyone who investigates in good faith, without accessing third parties' data or interrupting the service.
- Our full policy, with deadlines and safe harbour, is in Disclosure policy.
- We publish how to contact us at /.well-known/security.txt.
- We publish the vulnerabilities fixed in our products, with what to do, in Security advisories (also in CSAF 2.0 format).