The only reason I didn’t become a hacker as a teenager, I think, was my morality. And, let’s be honest, being danger averse. That was before I learned that hacking mostly means understanding how systems work. Taking a thing apart piece by piece and putting it back together, until you’re confident you can improve it if you need to, or break it if you want to. By that definition, I became one.
One of the people who got me there is Adrianus Warmenhoven. I met him at BarCamp Yerevan when I was a student, and he was one of the people we looked up to, a reference that what we were doing could go somewhere. He built one of the first internet businesses in the Netherlands in 1993. He ran a free ISP for ING in the late nineties with a crew of fifteen, and put a satellite internet link into Tanzania. Then came decades in security: evangelist at Bitdefender, CISO at Tesorion, defensive strategist at Nord Security. He holds around ten US patents and taught a university class literally called Hack Everything. One recommendation on his profile says you may not agree with this man, but if you don’t at least listen, it will be your detriment.
So we listened. I came in grumpy about the state of things and hoping we’d find a more positive outlook together. We did, eventually. Getting there took a detour through some dark places.
The same old bugs
Adrianus recently joined the Guild of the Grumpy Old Hackers, a group of peers, mostly Gen X. Most of them are really affable people. The grumpiness comes from watching the same mistakes get redone in every hype cycle. New technology arrives, and the same old bugs and the same old vulnerabilities come back with it.
I asked whether AI is any different. His answer started with the internet itself. It was built to share information. Telnet, almost no security, no firewalls, and it went fine for a couple of years. Then people started making money with it, and the moment money is involved, there are people trying to get some of it in less nice ways. So security was added afterwards: SSH, SSL, TLS. It never slowed development down.
The same thing is happening with AI. He compares it to the arrival of home improvement stores. Everyone thought it was the end of contractors, because now anyone could go and get the wood. What actually happened was more accidents, a lot of shoddy work, and the contractors are still in business. Vibe coding an app that organizes your files is fine. A service for other people brings resource starvation, lock congestion and memory pressure, terms nobody has heard of when they start vibe coding. So they don’t mention them to their AI, and the AI doesn’t take them into account.
The engineer who said no
This is the part of the conversation I didn’t see coming.
Adrianus calls AI the most inefficient programming language humanity has ever invented, and also the most flexible one. When I asked why it’s being abused so much, he pointed at something I had never thought about. My “problem” with AI is not really a technological one. It’s an ethical one.
Until a couple of years ago, anyone who wanted to build software that manipulates people, tracks everything and squeezes money out of them had to find like-minded developers first. That was a small barrier, and it held back a lot of people, because a lot of engineers are pretty ethical. If you found one, it meant a lot of other people didn’t find that one.
Now they don’t need anybody. They can sit down for a week, have servers running, and nobody will tell them “we shouldn’t do this” or “we shouldn’t track somebody.”
“There’s no backtalk on any of the bad moral decisions.”
The morality of the engineer was a barrier nobody designed. And it’s gone.
Nobody cares about your local harness
I pushed back with the question a lot of engineers are asking. Doing things right usually means doing them slowly, and the local models don’t match the speed and quality of the proprietary ones. So what can people do?
“I’m not happy to give this answer, but the answer is not a lot.” In the old days, a web server behind Cloudflare put him on equal footing with any enterprise online. With AI, there is nothing a person or a small company can buy or rent that does the same against racks and racks of H200s.
He thinks the current pricing is a startup strategy with an extremely large venture capitalist behind it: shape the field so competitors can’t follow, the way Netflix did. And the target isn’t us. It’s large enterprises, governments and militaries, hooked on one ecosystem the way they got hooked on Microsoft Office. If the two of us ran a local model and earned 10,000 euros a month with it, he says, they couldn’t care less.
Crickets
I used to tell people not to share their birthday online. Now people hand their mental health, where they live and what they eat to companies that are a few years old. Adrianus is beyond worried. He’s flabbergasted. A leak of ten thousand records from a company makes the news and people get angry. A week later there’s an article about 100,000 leaked AI chats. Crickets.
His explanation is the “does this benefit me” equation. The risk isn’t immediate and you can’t see it. The benefit, the answer from the AI, is right there. And from his work on NordVPN’s security advisory board he knows the risk doesn’t expire. Aggregated data sets on the dark web contain small nuggets about him that haven’t been online for ten or fifteen years. Once it’s online, it finds its way to your profile, and deleting it later doesn’t matter anymore.
Then he talked about undress apps, and about what it takes to build one, train it, and ship it, and why nobody is picking up pitchforks. I’ll let you hear that part in his words.
Locks buy you time
I dug out an article Adrianus wrote in 2015, where he named abundance as the root cause of security problems: we push out far more code than there are people who can secure it. I asked whether security is now catching up with the amount of code.
On the defender side, he says, there are leaps and bounds. Scanning code and reasoning about it got much faster. But he also ran a few tests on the money side, and by his estimate it’s now five to seven thousand times cheaper to produce new code than to review it. For most code, unless it’s critical, nobody will pay for that review, not even with AI.
So we’re where physical locks are. A lock never promises that no burglar gets in. Read the rating and it says how long it resists an attack. Half an hour, maybe. When a former chief of the Dutch intelligence service started a cybersecurity company, one of the first things he did was buy cyber insurance. Adrianus took that as telling. Someone with that much inside knowledge doesn’t believe you can defend everything either. Security is now partly an insurance question, and you pick your battles: what is critical, and what are you willing to let break.
It’s just data
This was the part that shifted something for me. Targeting used to start with a company: they’ve got money, let’s try to hack them. Now it runs the other way. It starts with a cookie. An ad network sees the same cookie on website A and website B. A data leak puts your email address on a dating site. The profile grows, gets sold on the dark web, enriched, and sold again, until it’s big enough to point at places where your data is vulnerable. Then it gets picked automatically.
“You will be automatically selected without the criminal ever knowing anything really about you. It’s just data.”
I called it a hackability index. Once you reach critical mass, you just get hacked. He liked the term, and then explained how the Shodan part of it works.
The opposite of a zip file
I grew up with the internet of the nineties, a place a million times bigger than a library, where you could reach people who knew things and actually talk to them. The day before we recorded, I saw a musician who is entirely AI-generated, with a Spotify account and thousands of followers. So I asked whether the internet is dying.
In its current form, maybe. Adrianus described the report pipeline we’ve all seen. Someone has a few facts, asks the AI for a nice one-page report with diagrams, and sends it to their manager. The manager doesn’t want to read a full page and asks the AI to summarize it. We’re doing the opposite of zip files.
He also watches Reddit, Hacker News and Quora with his own bot detectors. The “your pet’s name plus the last thing you ate” posts we’ve all scrolled past are engagement bait, there to get more information out of people. Then he took it somewhere I didn’t expect: a 15-minute bike tire repair, and why skipping it explains the subscription economy. If he wrote his 2015 article again today, he says, the root cause would be convenience.
Iron Man in the cave
There is a positive outlook, and Adrianus didn’t make me wait too long for it. LLMs might be the final interface between people and computers. He already codes by voice sometimes. And outside the corporate sphere, people are building again. The Internet Archive takes in about a petabyte every ten days and stores it the old-school engineering way. And this year’s DEF CON badge carried bunnie Huang’s open source chip, which anyone can download, inspect and build on, in the hands of thousands of attendees.
That’s why he’s cautiously optimistic. Cautiously, because most of humanity will stay with the short videos and the dopamine. His advice is not to look at them, and to look at the hacker communities instead. That’s where people are figuring out how to build things themselves, how to protect the people they love, and what happens if the internet breaks.
The difference between a hacker and a security professional, in his talks, comes down to two Marvel pictures: Iron Man in the cave and War Machine. Which one he identifies with, and why the security boom didn’t produce more of the first, is worth hearing in his own words.
And when I asked how he lives with knowing how all of this works, the answer was simpler than I expected. You’ll hear it near the end.
He also has book recommendations: one he calls not very positive but important, which most people only know from the movie, and the cyberpunk he thinks people have forgotten the meaning of. Markus’s question from the previous episode, about what grounds you, got an answer I recognized. His question for the next guest is on the cover of this episode.
Find Adrianus at warmenhoven.co, on LinkedIn, and if you’re serious, on MentorCruise. In his own words: friendly, yes. Easy, no.
