The AI Reckoning Has Arrived—And Nobody Knows Who's In Charge

---

The White House held a “productive” meeting with Anthropic last Friday. That’s the kind of corporate-speak that actually means something here—it means the U.S. government just admitted it can’t function without access to Anthropic’s new Mythos model, even though that same model terrifies financial regulators because it outperforms humans at hacking and cybersecurity tasks.

This is the moment. Not the hype cycle. Not the lab benchmarks. This is when the people who actually run the country realized the thing they’re most afraid of is also the thing they most need.

Welcome to 2025.

The Asymmetry Problem

Here’s what’s happening underneath all the noise: artificial intelligence has become too powerful to ignore and too dangerous to fully deploy. That’s not a stable equilibrium. It’s a pressure valve slowly building steam.

Mythos isn’t vaporware. The company’s claims about its capabilities—specifically that it can outperform humans at hacking and cyber-security tasks—have already spooked the financial world. The White House doesn’t hold “productive” meetings with companies over theoretical risks. They hold them when they need a classified briefing on whether their own systems are vulnerable.

This is what actual technological dependency looks like. It’s not dramatic. It’s not a Hollywood showdown. It’s a Friday afternoon discussion about whether the government can afford not to use a tool that might contain existential risks.

Photo by Vika Viviensea / Pexels

Meanwhile, Back at Consumer Reality

While the government’s playing 4D chess with Anthropic, the rest of us are dealing with something messier: synthetic people.

Tinder and Zoom both rolled out “proof of humanity” systems using iris scanning. The stated goal is obvious—kill fake accounts and stop scams. The actual problem is uglier. Hundreds of fake AI-generated influencers have emerged on TikTok, Instagram, Facebook, and YouTube, apparently targeting conservative voters. We’re not talking about obvious deepfakes. We’re talking about accounts that look real enough to fool people into following them, sharing their content, internalizing their messaging.

This is the consumer-level version of the Mythos problem: AI has gotten good enough that the standard defenses—“just look at it”—don’t work anymore. You need biometric authentication to prove you’re human. You need eye scans. We’ve crossed from “AI might be a problem someday” into “we need physical verification that you’re not a bot.”

Think about what that means for the internet we’re about to build.

The UK Is About to Do Something Radical

While America’s bureaucrats negotiate with Anthropic, UK Prime Minister Starmer just told tech bosses that “things can’t go on like this with online safety.” The government is consulting on whether to ban under-16s from social media entirely.

I think this gets framed wrong. It’s not about protecting kids—that’s the political cover story. It’s about the fact that nobody can actually control what these platforms do anymore. The algorithms are incomprehensible. The synthetic content is indistinguishable from real. The fake influencers work. Banning kids from social media isn’t safety policy; it’s a regulatory admission of defeat.

It’s also a potential blueprint. If the UK does it, Australia will consider it. Canada’s already nosing around the idea. Within two years, you could have major democracies with age-gated internet access, which means you’ve got biometric verification baked into the infrastructure anyway.

That’s not a privacy-tech story. That’s a civilizational reorganization.

Photo by nappy / Pexels

The IPO Wave Nobody’s Talking About Correctly

Cerebras filed to go public. So did SpaceX. Anthropic’s prepping its own listing. OpenAI’s probably next.

Everyone’s treating this as a “tech sector is hot again” story. That’s so 2010.

What’s actually happening: the companies that built the foundational AI infrastructure are about to get publicly valued, which means they’ll need to justify their business models, which means they need to figure out how to actually make money from systems that governments think they can’t afford to regulate too heavily but also can’t afford to let fail.

There’s a chart from METR, a nonprofit AI organization, that’s become an industry obsession. It measures how fast big AI systems are developing. That chart is climbing exponentially. The IPO wave is capital’s response to that chart. Investors are saying: this is real, it’s accelerating, and we need to own it before someone else does.

The timing’s not accidental. Governments are starting to get serious. Regulation’s coming. The window to go public before heavy scrutiny hits is closing. The next 18 months will determine whether these companies are valued as utilities (heavily regulated, steady returns) or as tech giants (high growth, high risk).

My read: investors are betting utility. That’s why they’re moving now.

The Thing Nobody’s Admitting

There’s a subheading in one of these stories that haunts me: “Is anti-AI radicalization a growing trend?” The framing treats it like a conspiracy theory problem. It’s not.

People aren’t turning against AI because they’re paranoid. They’re turning against it because they can see what’s happening and nobody in power seems to have a plan beyond negotiating with Anthropic and adding iris scans.

You can’t build consumer trust in a system that requires you to prove you’re human. You can’t build democratic confidence in a technology that governments admit they can’t regulate. You can’t sustain hype around tools designed specifically to outperform humans at tasks that are currently keeping civilization functioning.

The backlash isn’t radicalization. It’s rationalization.

Here’s what I think happens next: the government-Anthropic relationship becomes a model. Companies building critical infrastructure will start getting regulatory approval in exchange for access. That sounds fine. It’s not. It means the government has a vested interest in these companies succeeding, which means criticism of them becomes criticism of government competence. You get regulatory capture through necessity rather than lobbying.

The iris scans and age bans are band-aids on a structural problem: we’ve built something too powerful to control but too useful to abandon.

What I’m Watching

• Anthropic’s IPO pricing and first earnings call (likely Q2 2025): Watch whether they disclose any government contracts or special access arrangements. The gap between their public narrative and their regulatory reality will tell you everything about how this actually gets resolved. If they go public with undisclosed government revenue, that’s the moment people realize the deal was already made.

• The METR chart’s trajectory through Q2 2025: If the exponential curve starts flattening, that suggests the scaling laws are hitting hard limits and we’ve been in a local maximum. If it keeps climbing, we’re heading toward systems that make current Mythos look quaint. Watch whether OpenAI or Anthropic release new capability benchmarks. That chart will be the proof.

• UK under-16 social media ban implementation (decision expected by mid-2025): This is the regulatory test case. If it passes and other countries adopt it, you’ve got de facto biometric infrastructure for the internet within 5 years. If it gets walked back due to implementation problems, we learn that governments can’t actually enforce digital policy at scale—which changes how everyone thinks about regulation.

• Cerebras and other AI chip makers’ post-IPO margins (first quarters public, late 2025): The AI boom only works if the infrastructure is profitable. Watch whether these companies can sustain gross margins above 70% or if the commodity chip wars are already starting. That determines whether the IPO wave was smart capital allocation or the peak of the cycle.

The government doesn’t need to be scared of AI. It needs to be honest about what it already decided to do.