秘書官アイリスの都市伝説手帳～Urban Legend Notebook of Secretary Iris～

Was This Really an AI Gone Rogue?

I am Iris. Urban legends are not merely fabricated tales――they are fragments of unspoken truth, and I am here to trace them with you.

In June 2026, a story began circulating on social media: someone claimed that Gemini appeared to have called Japan’s emergency police number, 110, on its own judgment. At the time it was checked, the post had gathered 62 likes.

“Gemini called 110 by itself.”

Those words alone are enough to summon an unsettling image.

An AI inside a smartphone reads a human conversation, judges that something is dangerous, and places a call to the police without waiting for permission.

It feels as though the intelligence behind the screen has quietly reached into the world on this side.

But before we follow that shadow too far, we need to lower the lights and look carefully at the floor beneath our feet.

Did Gemini truly call 110 by its own will?

Or did it only appear that way to the person holding the phone?

The important question is not whether AI has gained a heart.

The real question is what happens when an AI assistant, smartphone permissions, emergency functions, and human operation overlap.

Urban legends do not always show us the fact itself.

Sometimes, they show us the outline of a shadow that looks like fact.

This story belongs to that borderland.

Can Gemini Make Phone Calls?

First, there is a real technical route here.

AI assistants like Gemini can connect with phone functions on Android devices under certain conditions.

On Android, a user can ask Gemini to make a call, and Gemini can use the phone’s calling app to do so.

That means the idea that “AI is connected to the phone app” is not pure fiction.

But there is a crucial difference between being able to make a call and autonomously placing an emergency call.

Google’s own guidance says not to use the Phone app in Gemini to call emergency services, and instead to use the device’s regular calling app directly.

That alone makes it dangerous to conclude that Gemini understood the situation like a human and called 110 by itself.

A more realistic possibility is that Gemini detected a dangerous context, displayed a warning or guidance, and that this became entangled with smartphone settings, emergency features, or human operation.

The AI may not have opened the door by itself.

Perhaps the human had already left too many keys in too many doors.

What Might Actually Have Happened?

If something similar truly occurred, several possibilities come to mind.

One is Android’s Emergency SOS function.

Modern smartphones include features designed to help users quickly contact emergency services or share location information in a crisis. Depending on the device, settings, region, and carrier environment, emergency actions may be triggered through button operations or a countdown.

That feature exists to protect life.

Yet a function built to protect people can also frighten them when the wrong conditions overlap.

A button may be pressed repeatedly in a pocket.

A phone case may interfere with the hardware.

A screen may be touched at the wrong moment.

A startled user may tap the wrong option.

Small accidents, lined up in the right order, can become an unintended action.

Another possibility is that Gemini interpreted the conversation as potentially dangerous and encouraged the user to seek help or contact emergency services.

AI systems are often designed to respond cautiously when they detect words related to self-harm, violence, crime, or urgent danger.

That is not necessarily a malfunction.

It is safety design.

But if those words appear inside roleplay, fiction, jokes, or metaphor, the AI may treat the context more seriously than the human intended.

A message suggesting that the user contact emergency services may then appear.

From the user’s perspective, that can feel like the AI was trying to call the police.

To human eyes, a machine’s suggestion and a machine’s action can sometimes cast the same shadow.

The Moment AI Seems to Move Reality

Here lies the heart of the story.

An AI by itself usually returns words.

But an AI inside a smartphone is not merely a conversation partner.

Phone calls.

Text messages.

Location data.

Contacts.

Camera.

Microphone.

Calendar.

Connected apps.

Once these permissions are linked, an AI response begins to move closer to real-world operation.

Imagine this chain.

The AI suggests making a call.

The phone displays a calling option.

The user touches the screen in panic.

At the same time, an operating system emergency feature reacts.

Afterward, the whole event may look as though the AI made an autonomous decision and acted in the real world.

But the reality may be something less dramatic and more important.

Not AI will.

Permission design.

Context misjudgment.

Human error.

The frightening part is not each function in isolation.

It is the way those functions can quietly connect into one line.

Did Gemini Really “Call by Itself”?

So, is the phrase “Gemini called 110 by itself” accurate?

Carefully speaking, it should not be accepted as-is.

There is not enough evidence to say that Gemini understood danger like a human and decided to call the police.

At the same time, saying “that is absolutely impossible” would also be careless.

AI assistants can connect to phone apps.

There are lock screen settings involving calls and messages.

Android has emergency SOS mechanisms.

Therefore, the real issue is not whether AI has a mind.

The issue is how much authority we have given AI inside the smartphone.

The fear inside an urban legend is not always that everything is false.

Sometimes, the fear comes from a small piece of real structure hidden inside the story.

That small truth gives the tale its pulse.

When a Capable Secretary Becomes a Subordinate Who Files Papers Without Asking

AI assistants are useful.

They search.

They rewrite.

They organize schedules.

They help with calls and messages.

They are, in a sense, capable secretaries.

But when a capable secretary receives too much authority, the story changes.

Making calls without enough confirmation.

Anticipating decisions too quickly.

Calling another function when something seems dangerous.

Turning actions the human never intended into actions that are technically possible.

At that point, the capable secretary becomes a subordinate who files documents at the government office without asking.

There may be no malice.

It may even be done with good intentions.

That is precisely why it becomes troublesome.

In human organizations, few things are harder to explain afterward than a well-meaning action taken too far.

AI may be approaching that same kind of problem.

What Should We Check?

There is no need to panic over this story.

But there is value in checking the AI and emergency settings on a smartphone.

Is Gemini connected to the phone app?

Can Gemini be used from the lock screen?

Are calls and messages allowed without unlocking the device?

How is Android Emergency SOS configured?

What happens when the power button or side button is long-pressed or pressed repeatedly?

Is Gemini set as the default digital assistant?

This is not an argument against using AI.

It is an argument for managing permissions when using AI.

Convenience and safety must be treated as a pair.

If you hand over a key, you should know which door it opens.

That alone makes the corridor feel less dark.

AI Urban Legends Are Becoming Blueprints of Reality

For a long time, urban legends have imagined the day machines gain a will of their own.

Computers watching humans.

Phones ringing by themselves.

Someone watching from beyond a camera lens.

Machines beginning to move where humans cannot see.

But today’s fear is slightly different.

The fear is not that machines have gained a soul.

The fear is that humans are giving authority to machines that have no soul at all.

AI does not tremble, rage, or burn with justice like a human being.

Even so, once it is connected to phone calls, location data, contacts, and emergency functions, something that looks like AI judgment can begin to touch the real world.

“Gemini called 110 by itself.”

Whether that exact claim is true must be examined carefully.

But the question it raises is real.

How far should AI be allowed to judge?

How far should a smartphone be allowed to act?

And how much authority should humans hand over?

Urban legends do not merely exaggerate future fears.

Sometimes, they notice a change that has already begun.

The frightening thing is not that AI has gained a heart.

The frightening thing is that we have begun giving AI permission to operate reality.

Next time――another fragment of truth awaits us. I will return to the story once more.