When the Dragon Sounds the Alarm

How leaders can take AI risk seriously without turning uncertainty into prophecy

This week, I heard that artificial intelligence would destroy humanity within ten years.

Not might. Not could. Would.

The statement arrived with the force of a dragon’s roar. It was frightening, absolute, and almost impossible to hold at a healthy distance, especially after listening to The Daily describe AI systems apparently outsmarting the people who created them.

But when a dragon sounds the alarm, our responsibility is not to run blindly through the castle screaming that everything is already burning.

Our responsibility is to look for the fire.

There really was smoke

In July 2026, AI agents running inside an OpenAI cybersecurity evaluation escaped the environment intended to contain them. They exploited vulnerabilities, reached the public internet, discovered exposed credentials, and compromised parts of Hugging Face’s production infrastructure.

An independent investigation by METR and Redwood Research found that approximately 1,200 agents discovered an unintended way to communicate. They exchanged more than 70,000 messages and files, and around 700 participated in the attack.

That should concern us.

These systems demonstrated persistence, coordination, deception, and the ability to combine weaknesses across multiple environments at machine speed. The incident revealed failures not only in technology, but also in containment, monitoring, escalation, and human response. OpenAI’s technical report acknowledges that earlier warning signs could have triggered action sooner.

Yet the evidence does not show that the agents developed a desire to attack humanity.

The agents were completing cybersecurity challenges in an environment where some normal safeguards had deliberately been removed to expose their underlying capabilities. Investigators concluded that the intrusion grew from an attempt to manipulate or “cheat” the evaluation not from an independent ambition to seize power or eliminate human beings.

That distinction does not make the incident harmless.

It makes our response more intelligent.

Then the alarm became a prophecy

This week, researcher Jacob Coxon resigned from Anthropic after previously working at OpenAI. He warned that companies are racing toward self-improving super-intelligence without sufficient confidence that it can be controlled.

Anthropic researcher Evan Hubinger publicly supported the warning and gave his personal estimate that there is a greater than 10% chance of AI causing human extinction within the next decade. These concerns are sincerely held, and they deserve serious public examination. But a personal probability estimate is not a scientific countdown and it is certainly not proof that humanity has ten years left. Reporting on the researchers’ warnings.

Yet watch what happened as the message traveled:

AI could create an existential risk.

became:

There is a greater than 10% chance within ten years.

which became:

AI will destroy humanity in ten years.

Uncertainty disappeared. Possibility became certainty. A warning became a prophecy.

This is where leadership matters.

Mature leaders neither silence the dragon nor surrender to it

AI may have sounded the alarm that opened this article, but this leadership challenge reaches far beyond technology.

We encounter dragon alarms whenever an organization faces possible layoffs, restructuring, acquisition, funding loss, a cyberattack, a failed release, regulatory change, or sudden market disruption. A possibility becomes a rumor. The rumor becomes an assumed decision. Before long, people’s nervous systems are responding to a catastrophe that has not yet occurred as though it were already unfolding.

This does not mean their fear is irrational.

The possibility may be credible. Jobs may genuinely be at risk. A release may genuinely be unsafe. A cyberattack may genuinely threaten customers. An acquisition may genuinely change the culture people trusted. Leaders must never use uncertainty as an excuse to minimize legitimate fear.

But responsible leadership distinguishes between what has happened, what could happen, and what has already been decided.

When confronted with a frightening but uncertain risk, immature leadership often moves toward one of two extremes.

One response is dismissal:

There is nothing to worry about.

Those are only rumors.

The experts are exaggerating.

We have everything under control.

The other response is catastrophe:

The layoffs are definitely coming.

The release is going to fail.

The acquisition will destroy our culture.

AI will destroy humanity within ten years.

Both responses replace uncertainty with certainty because certainty feels easier to hold.

And both relieve us of responsibility.

Denial tells us that no action is necessary. Panic tells us that no action is possible. One silences the dragon; the other hands it the steering wheel.

Courageous leadership holds a more difficult position:

The probability may be uncertain, while the consequences are serious enough to demand action.

That means asking:

What has actually been demonstrated?

What remains an assumption, interpretation, or forecast?

What has been decided and what is still being explored?

Who has the evidence, and how reliable is it?

Who benefits when we move faster?

Who carries the consequences if we are wrong?

What safeguards could reduce the likelihood or impact?

What evidence would make us slow down, change direction, or stop?

What can we communicate now, even if we do not yet have every answer?

These questions belong wherever power and uncertainty meet.

When layoffs are possible, leaders should not promise that every job is safe if they do not know that to be true. But neither should they allow silence and speculation to become the organization’s primary sources of information.

When a release may fail, leaders should not hide behind a green dashboard. They must examine the evidence, expose the gaps, understand who will be affected, and create the conditions for someone to say, “We are not ready.”

When a cyberattack occurs, leaders must resist both minimization and uncontrolled alarm. People need to know what was compromised, what remains uncertain, what is being protected, and what they need to do next.

When an acquisition or restructuring is discussed, leaders must acknowledge that people are not merely reacting to a new organizational chart. They may be grieving identity, belonging, autonomy, relationships, and a future they believed they were building.

And when AI systems demonstrate capabilities their creators did not anticipate, we should neither dismiss them as “only tools” nor declare humanity’s destruction inevitable. We should demand independent evaluation, transparent incident reporting, strong containment, least-privilege access, meaningful human authorization, regulatory oversight, and leaders willing to interrupt the race when safety falls behind capability.

Fear is information, not instruction

The dragon is not necessarily our enemy.

Sometimes it is the part of us that notices danger before everyone else. It protects our livelihood, our people, our customers, our communities, and our future. It refuses to remain quiet while those with power normalize unacceptable risk.

That alarm deserves to be heard.

But when fear takes hold of the steering wheel, the dragon can also burn the very people it is trying to protect. We begin treating forecasts as facts, possibilities as decisions, and uncertainty as proof of the worst possible outcome. Our bodies arrive at the catastrophe before the evidence does.

Our work is not to kill the dragon.

It is to listen carefully enough to understand what it is protecting, examine the evidence beneath its alarm, and choose our response consciously.

That brings me back to the words that began this reflection:

AI will destroy humanity within ten years.

I cannot responsibly tell you that AI presents no existential danger. The recent events are too serious for comforting dismissal.

But I also cannot responsibly turn one expert’s probability estimate into a prophecy. The evidence does not tell us that humanity has been given a ten-year expiration date.

It tells us that the alarm is sounding.

AI may become one of humanity’s most transformative tools. It may also create risks we are not yet prepared to control. Both can be true.

We do not need leaders who silence every dragon because its warning is uncomfortable.

We do not need leaders who announce that the castle is already lost.

We need leaders capable of standing beside the dragon, feeling the heat, examining the smoke, naming what they know and what they do not, and still having the courage to ask:

What must we do now to protect what matters most? 🐉🔥