The AI Reckoning Is Here—And It's Getting Messy

---

The White House called Anthropic to a meeting. Not an invite-only chat. A productive meeting. The kind where people sit around a table and talk about what happens if a company’s new AI model gets loose in the world.

That’s the moment you know the bloom is off the rose.

For the past two years, the AI boom has felt like one long victory lap. Better models every quarter. Billions flowing in from investors who’ll fund literally anything with “AI” slapped on it. VCs betting their entire thesis on the idea that the exponential curve goes up forever. But Anthropic’s new Mythos model—which the company itself claims can outperform humans at certain hacking and cybersecurity tasks—just yanked the emergency brake. Suddenly everyone’s worried. Financial institutions are spooked. Government officials are thinking about what “critical for security” actually means when a private company builds it.

This is what happens when technology moves faster than governance. Again.

Photo by Александр Ершков / Pexels

When Competence Becomes Dangerous

Here’s the thing about Mythos that’s got people nervous: it’s genuinely good at things that used to require actual expertise. Anthropic’s own framing—that this model can outperform humans at hacking tasks—is either a flex or a warning depending on who you ask. For a cybersecurity person in a bank, it’s a warning. For a nation-state looking at offensive capabilities, it’s a recruiting tool.

The White House meeting wasn’t a coincidence. U.S. officials believe Mythos could be “critical for security,” which is corporate speak for “we can’t afford to let another country own this.” It’s the exact moment when a tech company becomes strategically important in the way that nuclear facilities or chip fabs are strategically important. That’s why you get the government on the phone. That’s why you get a “productive” meeting instead of a press release.

My read: this is the end of the “move fast and break things” era for frontier AI. What’s coming next is messier. More regulation. More scrutiny. More meetings like the one Anthropic just had, except they’ll happen to every company building something that could theoretically go sideways. It won’t feel like progress. It’ll feel like bureaucracy catching up to innovation.

Meanwhile, Your Social Media Is Infested

While governments are worried about what AI could do, it’s already doing something simpler and uglier: manufacturing consent at scale.

Hundreds of fake pro-Trump avatars have flooded TikTok, Instagram, Facebook, and YouTube. Not bots posting spam. AI-generated fake influencers. Convincing enough that people follow them. Engaging enough that the algorithm promotes them. And according to reporting, they’re specifically designed to hook conservative voters. This isn’t a technical problem. It’s a political one wearing a technical disguise.

Tinder and Zoom are fighting back with iris-scanning technology to prove you’re actually human. Eye-scans. The kind of biometric surveillance we used to argue about in privacy circles. Now they’re a feature you opt into just to prove you’re not a deepfake. The defensive move has become so normal nobody even blinches.

This is the part of the AI revolution nobody on the VC circuit talks about at parties. It’s not the moonshot. It’s the wreckage. Fake accounts spiking. Malicious scams proliferating. Trust eroding in real time across every platform. And the solution? More surveillance. Iris scans. Proof of humanity. We’re building a panopticon to defend against an army of ghosts.

Photo by nappy / Pexels

The IPO Gold Rush Is Just Getting Started

Meanwhile, Cerebras just filed to go public. So did SpaceX. Anthropic and OpenAI are prepping their own offerings. This is what a wave looks like. Not a gentle tide. A wave.

When everyone who built something remotely connected to AI starts racing for the exit, it means two things. One: they think the public market is finally ready to price in the hype. Two: they’re cashing out before something breaks.

I’d bet on both.

The real tell is the METR chart. This is a nonprofit AI organization tracking how fast companies are building big AI systems—basically measuring the slope of the curve. And according to multiple sources, it’s become “an industrywide obsession.” When an obscure metric becomes an obsession, you’re watching an industry pray for a metric that justifies the valuations everyone’s committed to. The chart becomes the truth because the alternative—admitting you don’t know what any of this is worth—is unbearable.

Government Finally Wakes Up

The UK government is moving on online safety. Prime Minister Starmer told tech bosses something basically boils down to: “This can’t keep happening.” Specifically, the government is consulting on whether to ban under-16s from social media entirely. Not regulate it. Ban it.

That’s not a regulation. That’s a regime change.

Here’s what I think’s happening: governments around the world are watching Mythos, watching the fake avatars, watching trust dissolve on platforms, and realizing they’ve been asleep for a decade. They’re not going to slowly build frameworks. They’re going to react hard. Some bans will be blunt instruments. Some regulations will accidentally kill innovation. Some will actually work. But the era of “let the companies figure it out” is ending faster than anyone expected.

The Violent Part

There’s one more headline that’s haunting: “AI Backlash Turns Violent.” The question is whether anti-AI radicalization is a growing trend.

I don’t know if it is. Genuinely. There’s a difference between a few angry posts and an actual movement. But I do know this: when technology moves faster than society can metabolize it, people get scared. When people get scared, some of them get angry. When they get angry, some of them get dangerous. The AI boom has created genuine wealth and genuine concern in roughly equal measure, and we haven’t yet figured out what to do with that friction.

What I’m Watching

• Anthropic’s next move on Mythos distribution. Did the White House meeting result in actual restrictions or just vibes? Watch for whether the model ends up in the hands of US government agencies and whether that becomes a permanent arrangement. Deadline: Q2 2025.

• Whether iris-scan adoption becomes standard or novelty. If Tinder and Zoom’s proof-of-humanity systems catch on, we’re living in a different world than we were six months ago. If they flop, the problem just gets worse. Check user adoption rates by mid-2025.

• UK under-16 social media ban passage and enforcement. This is the first real government test case. Does it actually happen? Does it work? Does it get circumvented in three weeks? The outcome determines how aggressive other countries get. Expect draft legislation by summer.

• The IPO window closing. Watch Cerebras’s valuation and demand in its public offering. If it’s sluggish, the AI IPO wave deflates fast. If it’s frothy, we’re in late-stage bubble dynamics.

The AI boom isn’t ending. It’s just getting real. And real is always messier than hype.