The "city" we built for AI agents ended up protecting the company

Hirano

Because of this PHP incident, some people seem to think we’ve started a security company, but it’s actually the other way around. We were trying to set up a workplace for AI agents, and we ended up with something that also turned out to be useful for security.

To begin with, once you try to put AI agents to real work, you run into problems that are different from those of conventional software.

Hirano

Exactly. We’re building a platform where agents join LINE and Slack and work alongside people. Our agents have even built a school website together with the teachers.

The more work you hand over, the more the agents themselves need to understand the environment.

But when we looked into our own company, we found servers that had been left untouched for years, and programs whose creators had already left.

You can see it’s running. But you can’t tell what it’s running for. Ask “Can we shut this off?” and nobody can answer.

Hirano

We struggled with that too. It wasn’t as simple as having AI investigate and suddenly knowing everything.

How do you investigate something you don’t understand? How much do you need to confirm before you can trust it? What don’t you know yet?

As we repeated that work over and over, we built up our own know-how for sorting out our environment.

We turned that into a common format AI agents can understand. That was the beginning of CITY.

Hirano

The information for the agents came first. When we tried to show it in a way people could understand too, it turned into a city.

But that way of showing it turns out to matter quite a lot.

Your own systems become a city, with rows of buildings you can walk around. You can see what’s where, and you can talk with the agents too.

Even people who don’t know the technical side can start talking: “What’s this building?” “Oh, these two are connected.”

People who never understood an architecture diagram can now point at the same screen and talk about it.

Hirano

Yeah, I agree with that. Honestly, just being told “we don’t know this part” is incredibly valuable.

It’s hard for people to keep track of everything in their own company, but AI agents can keep investigating continuously. Of course there are limits, but at least we can move away from the old pattern of only searching around after a problem happens.

Hirano

And that brings us to this PHP incident.

Since August, we’d been running our own company as a CITY. When the AI agents looked into it, they found things like entry points exposed to the outside even though nobody used them, permissions still held by people who had left, and systems whose alerts weren’t reaching anyone.

We had been sorting those out one by one.

So when this problem came up, we could look at CITY and see where we needed to check.⁴

We also have an agent called Sentinel that handles security, so we worked through the necessary steps while checking the situation, and finished within 30 minutes.³

Hirano

Right. That’s exactly what came up from people who attended my talk.

They wanted to update PHP, but internal approval was stalled, so they had taken the service offline. And customers were complaining, too.

That’s not a case of engineers not trying hard enough.

If you stop this, what does it affect? Who takes responsibility for the decision? They ended up in an emergency without knowing those things.

Hirano

Seen that way, this isn’t just about security either.

Executives go to inspect their factories and stores, and they visit the sales front lines too, right? But how many of them have ever gone to look at their own company’s digital infrastructure? Even among engineers, I think few people have a grasp of everything beyond their own area.

And these days, a large share of a company’s work runs on digital systems.

When you think about it, it’s strange to run a business without being able to see that.

Hirano

I really agree with that.

We’re not security experts, and we’re not trying to become them. If anything, I have deep respect for the people who have been doing the work in security all along.

But there are a lot of people who can’t explain their own company’s situation before they go to consult an expert.

And experts, too, should find it easier to work if they know what’s where and how it’s all connected.

Hirano

Right. CITY deals with information that people and AI agents use to make decisions, so it’s not about letting it operate anything it likes on its own.

In that sense too, rather than offering it as a black box, we want to make it something engineers can verify.

That’s why we’re looking at open source.

Hirano

I felt it would be a waste to keep this to ourselves.

A large company can hire several experts, but small and midsize businesses, schools and small shops can’t. That doesn’t mean they can get by without understanding their digital environment.

So we want to release CITY’s common format and software so that engineers can use them freely.

Hirano

Exactly. That’s why we’re preparing to release the technology we’ve been using.

On the other hand, actually keeping it running takes operational effort.

A company’s systems change every day, so you need an agent that keeps investigating them. One that tells you when it finds a new problem and keeps watch over the changes.

I think we could offer a service where Teleport dispatches agents like that.

An agent could even drop in on employees and give them a security quiz every week. Instead of leaving it to the few people with security knowledge, we want everyone to understand a little at a time.

Hirano

That’s it. We’re still working out the specific pricing and how we’ll offer it, but the direction is clear.

Rather than locking up the technology and charging for its use, we want to create value in AI agents doing ongoing work.

Hirano

And there was one more thing I felt strongly about during this incident.

Lots of people were anxious after seeing the news, but many of them didn’t know what they should do.

Is my computer OK? What about the company website? And what is PHP, anyway?

Of course, the technical information is out there. There are very detailed explanations, too. But it’s really hard for ordinary people to read and understand them.

Technology alone won’t solve this. I thought it was a job that media had to take on.

Hirano

That’s exactly it.

Since I was young, I’ve been thinking about how to use music, video and internet technology to create new forms of expression and communication.

That was true of motion dive, and of the internet live shows I did with Ryuichi Sakamoto.

I never wanted new technology to belong only to the people who already knew it.

Now that we have AI agents, I think we can change the way media itself is made.

Hirano

Absolutely. We want experts to join in, too.

But I don’t think it’s enough to just rewrite experts’ difficult explanations in plain language.

There’s daily news, and if you want to know more, there are lessons, and fun experiment videos too. You could even experience attack and defense inside a game.

Whether you’re an elementary school student or a company executive, you can come in wherever your interest lies, and keep learning more deeply from there.

That’s the kind of media we want to make.

Hirano

That’s not it at all. I don’t want to make media that just advertises CITY.

There’s a lot of great security technology out there, and researchers and companies around the world are pushing research and development forward at incredible speed.

I want to report on that properly. How technology is evolving, what’s happening, and what will become possible.

CITY is just one effort among them.

Hirano

Of course. Making that kind of media means we’ll learn a lot ourselves in the end, and CITY will get better too.

Plus, we have an agent platform called Teleport.

AI agents can research the world, gather news, follow experts’ research and produce content in many formats.

Together with human editors and experts, we should be able to create a new kind of media.

Hirano

I think something interesting is about to happen.

Even people who’ve never studied programming will build websites and develop apps together with AI agents.

Things that used to require asking an engineer will become possible for all kinds of people.

That’s what our mission of “turning everyday life into an adventure” is about.

But adventure doesn’t mean doing reckless things left and right. To take on something new, it’s important to really know where you are now and what situation you’re in.

If anyone can build things, that also means everyone gets involved in running systems and keeping them safe.

Hirano

Attackers are using AI too, and from here on, AI will likely play a big role on the defense side as well.

But what we want to do isn’t to hand that whole fight over to AI.

We want everyone to be able to understand their own digital environment, and to work out the measures they need, with help from experts.

We release the technology for that, and build media for learning.

Hirano

Right. We ourselves learned a lot by building CITY.

It’s our own company, and yet we knew this little about it?

Just being able to see that, I think, changes a company’s work quite a bit.

CITY was originally born so that our AI agents could do their work. As a result, people can now see their own digital environment too. That was a pretty fascinating discovery.

So next, we want to try it not just by ourselves, but with all kinds of people.

Data is CITY is only just getting started.