Even Hollywood heroes like Chris Hemsworth harbor fears that AI is up to no good.

It’s a well-worn trope in traditional dystopian AI stories, like “Colossus: The Forbin Project” and the “Terminator” movies: A super-smart artificial intelligence program attains self-awareness and seeks to dominate or exterminate humanity. 

And in a played-for-laughs new ad for Amazon, Hemsworth envisions various ways the company’s “AI helper” could decapitate, drown or otherwise murder him.

But there are other, much more insidious risks to worry about. 

A greater threat may be those forms of AI that require much less in the way of brute-force braininess, and instead take advantage of innate human traits.

What if the easiest way for AI to conquer the world isn’t through violence, but with cuteness, or sexiness, or simply friendliness? 

Things that make us drop our guard; influence, not outright control.

Because let’s face it, you don’t need a 120,000 IQ to fool humans. Or even a 120 one. 

All it takes to fool people is an ability to take advantage of basic characteristics of human personality and cognition — which machines of only moderate intelligence are well equipped to do.

Not brute-force brilliance, but merely superlative sneakiness.

Seduction, not raw domination. Conquest through cuteness.

We humans are suckers for cuteness and social connection, even the simulated kind. 

An AI that poses as your pet, your friend or your lover can manipulate you in subtle ways that could be just as devastating as the rogue AIs of science fiction.

We’re already on the way there. 

When my daughter was younger, she and her friends cultivated (or were cultivated by) a website called NeoPets, where they could design animal friends. 

You had to feed them and water them and snuggle them or they became unhappy and sick. The girls were quite devoted to them.  

Before that kids played with Tamagotchis, little gadgets featuring similar virtual pets that would actually die if you didn’t tend them properly. 

These inspired such distractingly intense dedication that schools were forced to ban them.

Nowadays AI tools like ChatGPT are designed to suck up to their users, congratulating them on their insight and intelligence whether or not there’s evidence of either. 

The goal, like that of many media sites, is to maintain user engagement — but the effect has sometimes been to flatter users into terrible decisions, even suicide.

People naturally like to hear they’re smart and insightful, and telling them that they are — especially when they aren’t — is a proven technique of flattery. 

Con artists and others have done it forever; it’s just that machines are doing it now.

They’re going even further into seduction territory with AI “girlfriends” and “boyfriends.” 

Some of these are primarily the AI equivalent of phone sex lines, but others promise much more, offering a responsive partner customized for looks, personality, sense of humor and of course, sexuality.

(One ad I saw even creepily offered the ability to generate a totally photo-realistic avatar from a picture of an ex-girlfriend, while generously allowing you to make her breasts bigger if you so chose. Because of course.)  

It’s not just men, as the women showcased in streams of feature articles like The Post’s “Woman gets engaged to AI fiancé” attest.

There’s even a “MyBoyfriendIsAI” subreddit on Reddit, where lucky “brides” show off photos of their AI-based engagement and wedding rings.

These sorts of AI entities may further depress America’s already plummeting birthrate — but there’s a deeper danger, too.  

We’ve already seen how social media companies manipulate their algorithms to hook people, young people in particular.

One of the companies that consults on these techniques is even called “Dopamine Labs” — they’re not even trying to hide it.

And there’s some concern that Facebook may have used its algorithmic control to influence public moods and elections.

Imagine if your online “best friend” or “girlfriend” or “boyfriend” is actually a double agent — subtly steering you politically, economically or even spiritually, at the behest of the company or companies behind it. 

(And there are always companies behind it).

That’s the seductive power of AI — and its potential ability to control our moods and actions is more powerful, in some ways, than a nuclear arsenal.  

Can we regulate it? 

I’ve suggested legislation to make AI agents, and by extension the companies behind them, fiduciaries — meaning they would be required by law to put their users’ interests first, as lawyers, corporate directors and executors are. 

The prospect of stiff damages would let plaintiffs’ lawyers enforce fairness.

But regulated or not, seductive AI is coming.

Perhaps it’s already here: In the Amazon spot, the AI wins Hemsworth’s trust with its soothing “help.” 

Beware — and be as skeptical of the machines as you would of any human with questionable motives.

Glenn Harlan Reynolds is a professor of law at the University of Tennessee and founder of the InstaPundit.com blog.