Hadrian co-founders Rogier Fischer, Olivier Beg and Maurice Clin — Image: Hadrian's co-founders. Credit: Hadrian, via Tech.eu.

Penetration testing used to be a yearly ritual for most companies. You booked a team, they spent a couple of weeks poking at your systems, and you got a long PDF. Then everyone moved on until next year.

That model is under pressure. On October 6, Amsterdam-based Hadrian raised $40 million to run penetration testing with AI agents that never stop working. Investors are betting that attackers using AI will force defenders to test far more often.

I coordinate software projects for clients in India and the USA, and security reviews are part of almost every launch we run. So I read this funding news closely. Below I explain what Hadrian does, what the numbers really say, and how agencies should think about their next pentest.

What Hadrian Announced

Here are the main facts from Hadrian's official funding announcement:

The round is $40 million. It was co-led by Forgepoint Capital International and SmartFin.
Existing investors joined again. HV Capital, Motive Partners, Picus Capital and Oetker Ventures took part.
Total funding is now $65 million. The money will go toward growth in Europe, the Middle East, Africa and the US, plus bigger engineering and research teams.
Customers include large brands. The company names McKesson, NBCUniversal, Francisco Partners, TotalEnergies, Amadeus, Leroy Merlin and Damen Shipyards.

Hadrian was founded in 2021 by Rogier Fischer, Olivier Beg and Maurice Clin. Forgepoint says Fischer and Beg met at 13 as young ethical hackers. The team now has more than 80 people across Amsterdam, the US and the UK.

SiliconANGLE's report on the round points out the timing. The funding came less than a month after an Anthropic disclosure. The AI company said criminal and state-linked groups had used its Claude models to automate attack chains and hunt for zero-day bugs.

How AI Penetration Testing Works at Hadrian

Hadrian sells two products that share information with each other.

Atlas maps a company's external attack surface around the clock. That means every domain, server, app and login page an outsider can reach. AI agents then check which of those exposures can actually be exploited.
Nova runs agentic penetration testing on demand. It goes deeper on specific targets, closer to what a human pentest team would do.

Because the two tools share context, a team can spot something odd in Atlas and send Nova in to dig further. They do not have to start a new engagement from scratch.

The company is clear that people stay in charge. In its words, humans set the mission and make the decisions that matter. The AI does the repetitive work in between, at a pace no human team can keep up all day and night.

CEO Rogier Fischer put it simply. He said "most people are focused on automating the wrong bits." His view is that the platform should copy the paths real hackers take, so customers can see their biggest risks first.

Hadrian press image announcing its $40 million funding round — Image: Hadrian's funding announcement graphic. Credit: Hadrian press release.

Why Penetration Testing Is Changing

Hadrian's pitch rests on three numbers from its announcement. Keep in mind these are the company's own figures, not independent research.

87% of organizations still run manual pentests. Attackers can now move at machine speed, but testing still happens at human speed.
More than 70% of security teams struggle to tell which exposures are really exploitable.
Only 0.47% of vulnerability scanner findings turned out to be truly exploitable in Hadrian's data. That means most alerts need no action at all.

That last point matches what we see on client projects. A scanner report with 600 "high" findings does not tell a small team where to start. People either panic or ignore the whole thing.

The attacker side is also speeding up. We covered one example in our post on AI cyberattacks on South Korean banks. When attackers can scan and adapt in hours, a once-a-year test leaves a long gap.

What the Customer Claims Say

Hadrian says customers have seen 10 times more visibility into critical exposures and 80% faster time to resolution. It also claims 5 times the return of manual pentesting. These are vendor claims, so treat them as a starting point for questions, not proof.

The customer quotes are more useful to me than the multiples. Hans Quivooij, CISO at Damen Shipyards, said the company used to rely on periodic assessments and a known asset list. Now it has "continuous visibility into what an attacker can actually see and potentially exploit."

That shift from a known asset list to what an outsider can find is the real story. Most breaches we hear about involve something the team forgot existed.

What AI Penetration Testing Can and Cannot Do

AI testing tools are good at some things and weak at others. It helps to be honest about both before you buy.

Where AI Agents Help

Coverage. They can watch every public asset every day, not just the ones listed in a scope document.
Speed. New subdomains, open ports and leaked keys get checked soon after they appear.
Noise reduction. Checking whether a flaw is exploitable cuts the list down to the issues worth fixing.
Retesting. After a fix, the same check can run again to confirm the gap is closed.

Where Humans Still Matter

Business logic. A tool may not know that a coupon code should never stack, or that one user should never see another user's invoices.
Context and judgment. Someone has to decide which risk matters most to this client this quarter.
Safe testing rules. Testing production systems needs clear limits, approvals and a way to stop fast.
Compliance sign-off. Many audits still expect a named tester and a formal report.

Hadrian itself frames the product as people plus agents, not agents alone. That is the right way to think about it.

What This Means for Agencies in India and the USA

Agencies sit in an awkward spot. We build and run apps for clients, but we rarely own the security budget. Yet if a client site gets hacked, our name often comes up first.

Here is how I see the change affecting our work:

Clients will ask about continuous testing. Larger buyers in the USA already ask for recent pentest reports during vendor reviews. Expect "how often do you test?" to become a standard question.
Launch-time testing is not enough. Many apps change every sprint. A test from six months ago says little about today's release.
The forgotten asset is the big risk. Old staging servers, test subdomains and demo apps often sit open long after a project ends.
Reporting rules are tight. In India, CERT-In requires certain cyber incidents to be reported within six hours. Finding issues early is far cheaper than reporting a breach.

Rules also shape how often you test. PCI DSS, for example, expects penetration testing at least once a year and after significant changes for systems that handle card data. Treat that as the minimum, not the goal.

A Simple Penetration Testing Plan for Your Agency

You do not need a large budget to improve. This is the plan I share with our delivery leads.

List every public asset. Domains, subdomains, APIs, admin panels and staging sites. Include the ones from closed projects.
Shut down what you do not need. The safest server is the one that is no longer online.
Run continuous external scans. Use an attack surface tool, AI-based or not, so new exposures show up within days.
Fix by real risk. Start with issues that are reachable and exploitable. CISA's Known Exploited Vulnerabilities catalog is a good filter for what attackers are using right now.
Book a human pentest for big releases. Payment flows, logins and anything with personal data deserve a deeper manual review.
Retest after every fix. Do not close a ticket until the check passes again.
Write down the rules of engagement. Agree on scope, timing, contacts and a stop button with the client before any test runs.

For the method behind this, NIST's technical guide to security testing (SP 800-115) is still a solid reference. It explains planning, testing and reporting in plain steps. Keeping software patched closes many of the gaps a test will find, as we explained in our guide to patch management after the FBI and Accenture breach.

Questions to Ask an AI Penetration Testing Vendor

How do you prove a finding is exploitable without harming live systems?
What can the agents do on their own, and what needs human approval?
How do you handle client data collected during testing?
Will your report satisfy our auditors and our clients' auditors?
How fast can we pause or stop a test?

How We Approach Security Testing at Debugged Pro

We keep an asset list for every client project and review it before each major release. When a project closes, we remove test environments and old credentials instead of leaving them online.

For high-risk features, we plan a focused test before launch and a retest after fixes. We also tell clients plainly what an automated scan covers and what it misses.

Not sure how exposed your product is? Our software development consulting team can review your setup and help you build a testing plan that fits your release cycle.

Key Takeaways

Hadrian raised $40 million on October 6, 2026, bringing its total funding to $65 million.
Its platform pairs nonstop attack surface mapping with on-demand agentic penetration testing through its Atlas and Nova products.
The company says 87% of organizations still test manually and only 0.47% of scanner findings are truly exploitable.
AI agents add speed and coverage, but humans are still needed for business logic, judgment and sign-off.
Agencies should move from yearly tests to continuous checks, plus deeper human testing for big releases.

About the Author

Nitin Kumar is the Lead Project Coordinator at Debugged Pro. He manages software and AI delivery projects for clients in India and the USA, with a focus on secure launches, practical testing plans and systems that teams can maintain after go-live.