Stop card testing, fake signups and bot attacks
Criminals now run these attacks with AI agents that change tactics when they're blocked. We find how they're getting into your site or app and close it.
Most fixes are settings and code in tools you already pay for, such as Stripe, Cloudflare, your email provider and your own app. We make the changes and test them, and we can keep watching afterwards.
1,240 payment attempts between 1am and 3am
Most were declined. Each one used a different card and a small amount, from addresses spread across several countries.
Being attacked right now?
These buy time while we work out how the bots got in. Some, like pausing free credits or revoking a key, also affect real users, so weigh that against what the attack is costing you.
Card testing on your checkout
Put a CAPTCHA or Cloudflare Turnstile on the payment step and rate-limit it. In Stripe, add Radar rules that cap cards per customer and new customers per IP, and tell your payment provider.
Fake signups or free-credit abuse
Pause free credits for new accounts and stop emailing addresses that haven't been verified. That caps the bill and protects your email reputation.
A leaked key or a breached supplier
Set a spending limit, issue a new key and revoke the old one. If a supplier was breached, replace every key and connection they hold: new one in first, old one out once your site works.
Logins being tried or taken over
Rate-limit the login page, reset passwords on affected accounts and turn on MFA for staff and admins. Check for changed payout details or email addresses.
Then tell us what you've seen: when it started, whether it's still happening and which tools you use. Please leave passwords, keys and card numbers out.
Get help with an attackWhat the bots are doing
The attacks most likely to hit an online business with signups, checkouts, APIs or AI features, and the signs you'd notice first.
Card testing
Bots run stolen card numbers through your checkout, donation form or saved-card signup to find the ones that still work, usually with small amounts.
Signs: A wall of declined payments overnight, or a pile of disputes a few weeks later.
Fix: A CAPTCHA on the payment step, limits per IP, device and card, payment provider rules and 3D Secure where it's needed.
Fake signups and free-credit abuse
Scripts create accounts in bulk to use free trials, AI credits and referral bonuses, or as cover for card testing. One founder logged 55,000 in a night.
Signs: Signups jump while activation drops, and the AI, SMS or hosting bill climbs.
Fix: Turnstile, disposable-email blocking, device checks, a card for trials and per-account usage caps.
The 55,000-signup post-mortem ↗Your emails going to spam
Every fake signup or test order sends a welcome, verification or receipt email to a junk or trap address.
Signs: Bounces and complaints rise, then real customers stop getting password resets and receipts.
Fix: Send nothing else until an address is verified, cap sends per IP, use a separate sending subdomain and suppress bounces.
Stolen API keys and breached suppliers
Scanners search public code, app bundles and AI tool configs for keys. Google has seen stolen AI access run up more than $100,000 of charges in a single day. Other attackers break into a software supplier and reuse its access to every customer connected to it.
Signs: An AI or cloud bill you don't recognise, or an email from a supplier saying they've been breached.
Fix: A list of the keys and connected apps each system uses, keys limited to what they need, spending caps and a rotation plan ready in advance.
Google on stolen AI access ↗Account takeover
Passwords leaked from other sites are tried against your login page at scale, with CAPTCHAs solved automatically.
Signs: Failed logins spike, and customers report orders, payout changes or emails they didn't make.
Fix: MFA, checks against known breached passwords, login rate limits and alerts on unusual sign-ins.
Prompt injection on chatbots and agents
Hidden instructions in a message, web page or file make a support bot or agent leak data or take actions it shouldn't.
Signs: Odd refunds, emails or data in the chat logs, or a model bill that doesn't match your traffic.
Fix: Tools limited to what each job needs, a person approving anything that sends or pays, and testing before launch.
Prompt injection testing ↗How we stop it
A fixed-fee check first, so you know where you stand before anything changes.
Check
With your written permission, we test signup, login, checkout, password reset, forms and APIs the way an attacker's agent would. We list the API keys and connected apps those systems use and check whether any have leaked.
Fix
We change the settings and code in the tools you already use, or give your developers a ranked list. Anything that could stop real customers paying is tested before it goes live.
Retest
We rerun the attacks against the changed system and check real customers can still sign up, log in and pay.
Watch
Alerts on signup floods, decline spikes, bounce spikes, leaked keys and changes to payment-page scripts. If a supplier you use is breached, you get a list of what to rotate, in order. What's watched, who gets alerts and our response hours are agreed before it starts.
What you get
A short report written for whoever has to act on it, whether that's your developers, your payment provider or your insurer.
- What we tested, when and with what permission
- How each attack could get in, with evidence
- Fixes ranked by risk and cost
- Retest results after the changes
- Alert settings, if you keep us watching
Why it's got worse
Attackers use AI agents too. Theirs rotate IP addresses, solve CAPTCHAs and change tactics when one gets blocked, so a single rule rarely holds for long. The volume has moved to many small attempts that are cheap to run.
Growth in AI-driven bot attacks in 2025. Bad bots were 40% of web traffic.
Thales Bad Bot Report ↗UK card fraud cases in 2025, while losses stayed broadly flat: more attempts for smaller amounts.
UK Finance ↗Who it's for
SaaS and AI products
Free trials, AI credits and public APIs are what bots farm first, and every fake account costs you in model and hosting bills. Stripe found AI startups faced up to 4.3 times the attempted fraud rate of startups overall.
Stripe on fraud at AI startups ↗Online shops and booking sites
Checkouts on Shopify, WooCommerce or your own build, plus booking and enquiry forms that bots fill with fake requests. We work with your platform's settings and the apps connected to it.
Apps built with AI tools
Apps made with Lovable, Bolt, Replit or a coding agent often go live without rate limits, signup checks or webhook validation, sometimes with keys in the front end. We close those gaps and can host and monitor the app afterwards.
Fix a vibe-coded app ↗We get attacked too
We build and run Raq.com, an AI platform with its own signups, payments and API keys, so we defend against these attacks on our own systems. Our agents audit our servers every day.
Vu has OpenAI Trusted Access for Cyber, which gives us specialist Codex security capabilities for code review. Findings are reproduced and checked before they go in your report.
Independent testing when you need it
If your insurer, bank or a large customer needs an accredited tester, send us the requirement. We'll arrange a suitable external assessment within the engagement and handle the fixes.
Penetration testing ↗Questions about AI fraud protection
Can you help while we're being attacked?
Yes. Tell us what you're seeing and which tools you use. We start with changes that slow the attack without blocking real customers, then work out how the bots got in and close that route.
Will blocking bots stop real customers or AI shopping agents?
It can if it's done bluntly, so we test real signups and payments after every change. Legitimate AI agents that shop or book for people are starting to identify themselves, for example through Cloudflare's Web Bot Auth, which Visa's and Mastercard's agent payment schemes use, so they can be let through while unidentified scripts are challenged.
Do we need to buy new fraud software?
Usually not at first. Payment providers, Cloudflare and most email services already include the controls that stop common attacks, and they're often switched off or left on defaults. If you need a specialist tool, we'll say which one and why.
A supplier we use has been breached. Can you help?
Yes. We work out which keys, tokens and connected apps that supplier could reach, rotate them in an order that doesn't break your site, and check whether any were used while exposed.
What access do you need?
For the check, your written permission and test accounts. For fixes, access to the specific tools involved, set up so it can be removed afterwards. We never need your customers' card numbers.
Can you work with our developers or agency?
Yes. We can hand them a ranked list of fixes, or make the changes ourselves and walk them through what changed.
How do you price the work?
The check is a fixed fee agreed before we start. Fixes are quoted from what the check finds, and ongoing monitoring is a monthly fee per site or app.
Tell us what's happening
A few lines is enough: what you've noticed, when it started and the tools involved, such as Stripe, Shopify, Cloudflare or your email provider. We'll reply with first steps and how we'd check the rest.
Please keep passwords, API keys and card numbers out of the first message.
Our security practices and certification ↗