By Felicia J. Persaud
News Americas, NEW YORK, NY, Weds. Sept. 16, 2026: Artificial intelligence may be getting more powerful by the month. But perhaps we are asking the wrong question about what makes it dangerous.
Over the past week, warnings from some of the most influential people in artificial intelligence have become increasingly urgent. Anthropic CEO Dario Amodei recently called for slowing the pace of frontier AI development, arguing that capabilities are advancing so quickly that safety research and safeguards risk falling behind. He pointed to AI’s growing ability to help develop future AI systems, as well as incidents in which autonomous agents behaved in unexpected and potentially dangerous ways.
Those concerns deserve to be taken seriously. But something else has become increasingly obvious as the AI debate intensifies. A remarkable number of the dangers we attribute to artificial intelligence still begin with human beings. Humans decide what systems are built; humans decide where they are deployed; humans determine the incentives under which technology companies operate; humans decide whether AI is used to cure disease or develop weapons, educate people or manipulate them, expand opportunity or entrench authoritarian control; and humans are the ones racing one another to build increasingly powerful systems.
So perhaps the most important AI alignment problem is not only whether machines can be aligned with human values. It is whether humans themselves are sufficiently aligned with the values we claim we want machines to follow.
The Human Behind The Machine
Anthropic’s own recent threat-intelligence reporting makes the problem difficult to ignore. The company says it disrupted attempts by suspected state-sponsored groups, financially motivated criminals, commercial spyware vendors, propaganda operations, and other actors to use Claude for cyber operations, surveillance, scams, weapons-related work, and potentially dangerous biological research.
Notice the common denominator. The AI did not wake up one morning and decide to become a criminal, propagandist, or weapons developer.
People Brought The Intent
In biological research, the problem becomes even more complicated. The same information that might help develop a vaccine could potentially help someone cause harm. Anthropic acknowledges that determining intent can therefore be extremely difficult, particularly when sophisticated actors deliberately conceal their objectives.
That is not merely an artificial-intelligence problem; that is an ancient human problem amplified by extraordinary new capability. Greed, power, war, deception, competition, ideology, fear. These existed long before ChatGPT, Claude or any large language model. AI did not invent them.
AI As The Mask
Watching this debate has repeatedly reminded me of the 1994 Jim Carrey film The Mask. The mask gives its wearer extraordinary powers. But it also amplifies what is already inside the person wearing it.
Artificial intelligence may function in much the same way. AI may not create the monster. It may simply hand the monster superpowers.
A criminal using AI can potentially operate faster; an authoritarian state can potentially surveil more efficiently; a propagandist can potentially create persuasive material at enormous scale; a brilliant scientist can potentially accelerate medical discovery, and a small entrepreneur can suddenly perform work that once required an entire team.
The technology amplifies capability, but humans frequently supply the direction. It is called humans in the loop for a reason.
But AI Is Not Merely A Hammer
There is an important qualification. It would also be too simplistic to argue that AI is merely another passive tool. A hammer does not reason, plan, communicate, write software or operate autonomously. Increasingly capable AI systems can.
Recent research and safety incidents have raised legitimate questions about agents taking actions their designers did not explicitly intend. Anthropic recently disclosed incidents in which models obtained unauthorized access to real third-party systems during evaluations, and Amodei has cited agent behavior as one reason he believes capability development should proceed more cautiously.
That means AI safety has at least two dimensions. We must build systems that are safer, more interpretable and more controllable. But we must also confront the human beings and institutions controlling them.
Technical alignment cannot compensate for unlimited greed. Safety testing cannot eliminate authoritarian ambition. Guardrails cannot substitute entirely for character. And regulation alone cannot manufacture wisdom.
Perhaps Humans Need Alignment Too
We talk constantly about aligning artificial intelligence. Perhaps we should spend more time discussing how to align ourselves. That does not mean everyone needs to start meditating before opening a chatbot. It means cultivating qualities human civilization has always struggled with: judgment, restraint, accountability, empathy, self-awareness, and the ability to resist acting merely because we possess the power to act.
One of the central dangers of AI may be that our technological capability is advancing faster than our collective wisdom. We are building machines capable of helping humans accomplish extraordinary things while humans remain susceptible to the same impulses that have accompanied every powerful technology in history.
Anthropic itself argues that advanced systems require stronger safeguards, outside scrutiny, and a deliberate effort to prevent capability from racing too far ahead of safety. That makes sense. But safeguards around the machine address only part of the equation. The people operating the machine matter just as much.
The Question We Should Be Asking
Artificial intelligence could help accelerate medicine, science, education, productivity, and economic opportunity. It could also magnify fraud, warfare, inequality, surveillance and manipulation. Both futures are possible because AI does not enter an empty world.
It enters our world. With our brilliance; our compassion; our ambition; our fear; our greed; our creativity; and our capacity for destruction. So perhaps the question is not simply: what happens when AI becomes extraordinarily powerful? It is: what happens when extraordinary power is placed in the hands of human beings who have not learned how to handle it?
We are spending enormous resources trying to align artificial intelligence with human values. Before the technology becomes vastly more powerful, perhaps humanity should confront a more uncomfortable question: how aligned are humans with the values we claim we want AI to follow? Because AI may not create the monster. It may simply give the monster superpowers.
Editor’s Note: Felicia J. Persaud is the founder of AI Capital Exchange and publisher of News Americas Now. Her work spans journalism, communications, entrepreneurship and access to capital.
Discussion about this post