Who Gets the Keys to Superintelligence? Zuckerberg's Access Control Problem

The Pitch: Superintelligence for Everyone

Last week, Mark Zuckerberg published a 6,500-word essay titled The Future is for Everyone: The Path to a Positive AI Future. It's not about a new model or benchmark. It's a manifesto about who should get access to superintelligent AI—and Zuckerberg's answer is: everyone.

He paints a picture where your AI agent knows your sleep patterns, watches your workouts, helps your kid bake, and works while you sleep. It's a teacher, a coach, a co-founder. He calls it 'Personal Superintelligence.' One person, backed by a whole team of AIs, could run a company that used to need fifty people.

That's the dream. But the essay's real argument is about access control: if superintelligence is dangerous, don't lock it up. Give it to everybody. Let a thousand AIs balance each other's power.

The Balance of Power Argument

Zuckerberg's logic is straightforward. If only one person has a superintelligent lawyer, that person wins every case, even unjust ones. The fix isn't to ban the lawyer—it's to give one to the other side. Same for cybersecurity: attackers get better AI, so defenders need equal firepower.

He calls this 'balance of power.' It's a market-based, democratic approach to AI safety. Instead of building one 'good AI' that decides for everyone—which he says is impossible because humans disagree on values—you let different people and institutions have their own AIs, with their own values and goals.

But Who Controls the Keys?

Here's where it gets messy. Zuckerberg's personal AI isn't open-source. It's trained by Meta, powered by Meta's servers, and delivered through Meta's apps and smart glasses. So when he talks about putting power in people's hands, he's really talking about renting it to them through his platform.

That's not access control—that's a lease. Meta decides what the AI can do, what data it can see, and when it can be updated. The user gets a taste of superintelligence, but Meta holds the master key.

A History of Broken Trust

Zuckerberg's essay leans heavily on Meta's founding mission: 'give people the power to build community.' But the company's track record on access control is, to be kind, spotty.

• Cambridge Analytica (2018): Third-party apps siphoned data from 50 million users without consent. That's an access control failure at the platform level.
• Myanmar (2018): An independent human rights audit found Facebook didn't do enough to stop its platform from being used to incite violence against the Rohingya. Content moderation—a form of access control—was too lax.
• Teen mental health (2024): A New Mexico court ordered Meta to pay $567 million and change product features for teens. Lawsuits continue over addictive design.

Each case shows the same pattern: when growth and engagement clash with user safety, the platform often chooses growth. Why would a superintelligent agent be different?

The Privacy Promise—and Its Limits

Zuckerberg does address privacy. He promises a 'completely private mode' where even Meta can't see your data. That sounds good. But the default mode is what matters. Most people never change defaults. And a personal AI that knows your health, finances, and family is a goldmine for advertisers—Meta's core business.

Even if the private mode is real, the AI itself is still a black box controlled by Meta. You don't know how it's trained, what it's optimized for, or who it really serves. That's not access control; that's delegation.

A Better Path: Open Access

If Zuckerberg is serious about giving everyone superintelligence, the answer is openness. Open-source models, local execution, and user-controlled data. Not a proprietary cloud service with a privacy toggle.

Open access isn't perfect—there are real risks of misuse. But it's the only way to genuinely distribute power. Otherwise, we're just swapping one gatekeeper for another, and 'the future is for everyone' becomes 'the future is for everyone who uses Meta.'

The Question We Should Ask

Zuckerberg's vision is seductive. Who wouldn't want a tireless, brilliant assistant that knows you better than you know yourself? But the essay's real topic isn't AI—it's control.

When we hand over our keys to a superintelligent agent, we need to ask: who's really in the driver's seat? If the answer is a corporation with a history of prioritizing engagement over ethics, then 'power to the people' is just a slogan.

Conclusion: Trust, but Verify

Maybe Zuckerberg has changed. Maybe Meta's next act is different. But trust isn't a feature you can ship in an update. It's earned through transparency, real user control, and a willingness to let go of the keys.

Until then, I'll keep my own keys. And I think you should, too.