One minute, someone is asking artificial intelligence what to cook with the chicken in the fridge.

The next, a headline warns that AI could wipe out humanity by the end of the decade.

We asked for an adobo recipe. How did we end up discussing the end of the world?

It sounds like a movie trailer. Except the warning came from someone who helped build the technology.

In September, Jacob Coxon, a former researcher at OpenAI and Anthropic, resigned and raised an alarming concern. AFP quoted him saying: “The people building AI earnestly believe that it could kill us all by the end of the decade.”

AFP also reported that Anthropic researcher Evan Hubinger supported the concern. He put his personal estimate of the risk above 10 percent within the next decade, while describing the risk from current models as low.

On October 5, Coxon took his concerns to the New York City Council. The Council’s official account confirms that he warned humanity could lose control of AI, potentially ending in human extinction.

Before anyone starts panic-buying canned goods, let us get this straight.

These are warnings and personal estimates, not established scientific probabilities.

“By the end of the decade” and “within the next decade” are different timelines. Neither is an expiration date stamped on humanity.

Ask what evidence supports the prediction, what could prevent it, and what remains unknown.

Still, imagine a mechanic handing back your car keys and saying, “Enjoy the drive. By the way, I’m worried about the brakes.”

You would ask questions before getting on the highway.

The question is what kind of danger we are talking about.

There is a difference between a person using AI to hurt someone and an AI system doing things people did not intend or cannot reliably stop. Treating them as the same threat makes for a scarier headline, but a weaker argument.

Imagine this hypothetical scene in a Filipino household.

A mother is preparing dinner when her phone rings.

“Mom, there’s been an accident. Please, send money now.”

It sounds like her daughter. She hears crying. An account number arrives by text.

She transfers the money.

Only afterward does she call her daughter’s usual number.

“Mom? I’m fine. What accident?”

The first voice was an AI-generated imitation.

This is not a reported incident. But Philippine authorities have warned about scam calls using AI-generated voices, including imitations of real people.

The cruelty is that the scam does not need the mother to be foolish. It needs her to be frightened.

It works by making love move faster than verification.

For that mother, the crisis does not wait until 2030. It happens before dinner.

That is human misuse: a scammer using a tool to exploit someone. It does not prove that a machine can independently wipe out humanity.

Losing control means something different: powerful systems acting outside human intentions, with safeguards that fail to keep up.

A system does not need to hate people to cause harm. What it can do, what it gets wrong, and what people allow it to access all matter.

Anthropic’s own safety-policy update describes models that can browse the web, run code, use computers, and carry out tasks involving several steps. It also describes safeguards, public risk reports, and plans for outside review.

Those efforts matter. So does checking whether they work. The people selling a technology should not be the only ones deciding whether it is safe.

If they get it wrong, the people paying the price may never have agreed to use it.

Even Donald Trump—who, as we know, rarely seems short on confidence—wants government agencies to check the cybersecurity risks of advanced AI. His June 2 executive order calls for assessing those risks and developing a voluntary framework for government access to certain models before release. Companies would have to agree to take part.

That does not mean he is preparing for the apocalypse. But asking “Is this safe?” seems fair before handing powerful technology the keys.

A voluntary check still leaves a question: what happens when a company says no?

Does the public get a say—or just the consequences?

Meanwhile, people have work, traffic, bills, and dinner. Must every Filipino also become a part-time detective?

As a journalist, I worry about fabrications passing for truth. As a mother, that imaginary phone call makes my stomach turn.

Yes, we should verify. Call back through a number we know. Check before sending money. Pause when someone pressures us to act immediately.

But public protection cannot begin and end with “be careful.” That is advice to the potential victim. What are we demanding of the people who build, sell, and deploy the tools?

Companies should tell us when we are dealing with AI. Dangerous uses should face independent safety checks. People harmed by these tools need a real person to contact, a way to challenge mistakes, and a clear path to hold those responsible accountable.

And whoever is supervising the machine must actually be able to intervene.

AI can suggest a recipe, explain an unfamiliar term, or help plan a trip. But being helpful in the kitchen does not prove a system is safe everywhere else.

And we should ask ourselves something uncomfortable: are we demanding protection, or accepting whatever comes next because the convenience is too good to give up?

Lest we forget, convenience does not cancel responsibility.

We do not know that AI will kill humanity by 2030. We should neither turn that deadline into a prophecy nor let uncertainty become an excuse to do nothing.

Let AI suggest what to cook. But who gave anyone permission to decide how much danger our families should accept?

We asked how to cook adobo. We never agreed to let anyone gamble with whether it would be our last.

Bulkang Kanlaon, nagbuga ng abo na umabot sa 300 metro ang taas