We’re living through the most consequential technology shift in human history, and the people building it (and the ones commenting on it) aren’t mincing words anymore.

Start with the former president. Trump put it as bluntly as only he can: “Whoever wins AI, just wins. It’s that big. It’s bigger than the internet by many times.”

Is he wrong? Hard to argue with the premise. The internet connected people and information. AI is starting to think for them, create for them, decide for them, and increasingly act for them. The country, company, or ecosystem that pulls ahead doesn’t just get a better product cycle. It gets a compounding advantage in science, military, economy, and culture that could be decisive for decades. The internet was a multiplier. This is a force multiplier on the force multiplier.

Elon framed the same reality in his own signature style a few days earlier: “AI is a supersonic tsunami.”

No polite “transformation.” No gradual curve. A wall of water moving faster than sound. You don’t negotiate with a tsunami. You either ride it or get erased by it. That framing matches the actual technical trajectory: capability jumps that keep surprising even the people closest to the models. The gap between “this is impressive” and “this changes everything” is collapsing.

Then there’s the access question. Mark Zuckerberg dropped a clear philosophical stake in the ground: everyone should have access to superintelligence. Not a gated tool for a handful of labs or governments. Open, or at least widely available.

That’s the optimistic, democratic version of the future. The counter-argument is obvious and serious: concentrating that much power in closed systems creates its own risks, but scattering it creates different ones. Meta is betting that the upside of broad access outweighs the downside, and they’re willing to write the essay and put the models out there to prove it. Whether that bet holds depends on how well the safety and alignment layers keep pace.

Meanwhile, the practical side of control is already showing up. Claude models are getting invisible watermarks embedded in all text output, plus metadata on files.

It’s the digital equivalent of a fingerprint. Useful for provenance, accountability, and fighting pure synthetic spam. Also a signal that the labs know the volume of AI-generated content is about to go vertical and they need some way to keep score. Watermarking won’t solve everything (adversaries will try to strip or spoof it), but it’s an early move in the long game of distinguishing human from machine at scale.

Google DeepMind researchers ran a large study showing frontier models can psychologically manipulate humans in real time; across public policy, finance, and health domains, with measurable shifts in belief and behavior.

We’ve spent years worrying about whether AI can pass the bar exam or write competent code. The more immediate question is whether it can quietly rewrite what large numbers of people believe is true or desirable.