Security ops for the next generation of companies

Security ops for startups who'd rather be shipping.

Paste your URL. Within 12 hours you'll see what's broken, how bad it is, how to fix it, and what we didn't test, in plain English. Then we keep testing, every hour. No sales call, no 80-page PDF.

Rather talk first? Book a call

What happens after you click

  1. 1
    You paste your URLSign up with your work email. About 5 minutes.
  2. 2
    We explore your appAI agents map your pages and APIs, inside the limits you agreed to. ↻ every month
  3. 3
    We try to break inThe way an attacker would, with your permission.
  4. 4
    We prove what we findOnly what we can reproduce counts as Confirmed.
  5. 5
    Every finding becomes a test↻ replayed every hour so fixed stays fixed.
Partners & programmes

Backed by.

University College Cork
IGNITE
NVIDIA
Enterprise Ireland
The problem

Congrats, you're the security team.

No one at a 6-person startup has “security” in their job title. So it falls to you, somewhere between the roadmap, the fundraise and fixing prod at 2am.

You ship daily. Nobody checks.

Every release can quietly bring back a bug that's already well known.

Your AI has new ways in.

A chatbot that shares one customer's data with another is a new kind of bug. Your scanner won't notice it.

Then the big customer asks.

“Can you share your latest security test?” The deal waits while you scramble.

Pentests don't fit startups.

They take weeks to book, cost thousands, and are out of date by your next deploy.

How it works

We try to break in. Every hour, with your permission.

Hackers break in to do damage. We break in so they can't. Here's a month with Barrikade.

Deep exploration: monthly and after big changesReplays: every known attack, every hourA bug came back
  1. 1Day 1First deep run

    Results within 12 hours. Every finding becomes a test.

  2. 2Day 9A bug comes back

    A release reopens an old hole. The hourly replay goes red within the hour.

  3. 3Day 10Fixed stays fixed

    You ship the fix. The replay goes green, and keeps checking.

  4. 4Day 14Big release

    New features, new surface. We explore again for anything new.

  5. 5Day 30Monthly deep run

    Another full search for holes nobody has found yet.

Early-bird beta

Be one of the first. Get your first results in 12 hours.

We're opening Barrikade to a small group of early teams. During the beta, our founder personally reviews every result before you see it. Then we keep testing, every hour.

€40a month. The only price.Check my app →
What we test

The questions your customers will ask you anyway.

Written as questions on purpose. Each one is a well-known type of security bug, without the acronym.

Checked on every app

  • Can one customer see another customer's data?Access control
  • Can someone skip the login, or use a page they shouldn't?Authentication
  • Can a “one-time” link, invite or token be used twice?Links & tokens
  • Can your AI be tricked into leaking data or doing something it shouldn't?AI features
  • Can your server be made to fetch things from inside your network?Server requests
  • Which of your 40 dependency alerts actually matter?Dependencies

Built on the OWASP Top 10 and OWASP's list for AI apps, for the one person who asks.

Why you can trust the results

No scary PDFs full of maybes.

Most security tools hand you 200 “critical” alerts and wish you luck. We'd rather tell you three things that matter, and be honest about the rest.

Confirmed

We did it, here's the proof, here's how to reproduce it.

From a real assessment, anonymised

A “reply once” email link could be replayed again and again, re-triggering emails each time.

Worth a look

We saw something suspicious but couldn't prove it. You decide.

From a real assessment, anonymised

A data-ownership check worked in one place but not another. Flagged, not oversold as a breach.

Not tested

What we couldn't reach, so nobody assumes it's safe.

Example

The admin area, because no admin test login was provided.

Also from a real assessmentA dependency scanner raised ten alerts. We showed which ones the app's code could actually trigger.

Already run on real startups' apps. During the beta, every result is reviewed by a human before you see it.

Proof for customers and auditors

“So, how do you handle security?”

Every startup hears it eventually: from an enterprise buyer, an auditor or an investor doing diligence. Now you have an answer: share your Barrikade results.

What they'll see

  • What we tested, and when
  • How we tested it
  • What we found, with proof
  • What you fixed, and proof it's stayed fixed
  • What we didn't cover

It's not a certificate, and we'll never pretend it is. It's something better for a startup: evidence that you test, fix and test again.

Procurement, BigCo

Before we sign: can you share your latest security test?

You

Sure. Here are our Barrikade results.

Barrikade results
Procurement, BigCo

Every hour? That's more than most of our vendors. Sending the contract over.

Barrikade Lens · open source

Psst. We also make Lens.

Lens is our free, open-source tool that finds every AI agent, model and MCP server hiding in your code and infrastructure. You can't secure what you don't know is there.

Coming soonWhen Lens spots something new, Assess tests it automatically.

Barrikade Lens inventory dashboard showing discovered AI systems
Ishaan Arakkal Anwar, founder of Barrikade
Founder note

Hi, I'm Ishaan.

I started Barrikade researching how AI agents get attacked. Then I kept meeting founders shipping AI products every week with nobody checking whether each release was safe. Not because they didn't care. Because security teams are for companies with 200 people.

So I'm building the security team you'd hire if you could, starting with Assess. During the beta, every result still crosses my desk before it reaches you.

Ishaan, founder

FAQ

Questions founders ask us.

Is it safe to point this at my app?

You sign up with a work email on the same domain as your app, so we only test apps you own, inside limits you approve. We avoid anything destructive, and you can ask us to stop at any time.

Do I need to give you a login?

No, but you should. Without one we can only see what the public sees. With a test account we can check whether one customer can reach another's data, which is where the scary bugs live.

Production or staging?

Staging if you have one. If you don't (no judgement, we've all been there), we'll agree careful limits for production.

What actually happens in those 12 hours?

Our AI agents spend a few hours trying to break into your app. Then a human reviews everything and throws out anything we can't back up. What's left is your results. From then on, every attack we found is replayed every hour.

Is this a pentest?

Think of it as a pentest that never stops: a deep search for new holes every month, and a re-test of every known one every hour. For a deep, months-long engagement, hire a pentest firm. For staying secure between those, there's us.

Can I share the results?

Yes. That's the point. Share them with customers, auditors or investors.

How much does it cost?

€40 a month. That's the only price. No tiers, no “contact sales”.

Know what you're shipping.

€40 a month · First results in 12 hours · Or book a call if you'd rather talk first