The AI Reckoning Is Here—And Nobody Knows the Rules Yet

---

The White House called Anthropic in for a meeting. Not a photo op. Not a listening tour. A “productive” conversation that ended with both sides pretending they’d reached some kind of understanding.

That’s the tell.

When your startup’s AI model is so powerful that the U.S. government needs to figure out what to do with it before it breaks something, you’ve crossed a line. Anthropic’s Mythos model—which the company claims can outperform humans at hacking and cybersecurity tasks—has spooked Washington enough to pull the company into a room and say, basically: we need this, but we’re terrified of it. The financial world is already nervous. Regulators are nervous. And Anthropic gets to walk out saying the meeting was “productive” while nobody actually explains what was decided.

This is what peak ambiguity looks like in 2025.

Photo by 🇻🇳🇻🇳Nguyễn Tiến Thịnh 🇻🇳🇻🇳 / Pexels

The Chaos Is Getting Too Visible to Ignore

You can’t run a society on fake people anymore. That’s the subtext of Tinder and Zoom rolling out iris-scan verification—“proof of humanity” tools that sound like something from a Philip K. Dick story but are actually just desperation wearing a tech costume. Fake accounts. Malicious scams. Hundreds of AI-generated fake avatars flooding TikTok and Instagram pretending to be pro-Trump influencers to hook conservative voters. The fakeness has metastasized so completely that the platforms can’t even pretend it’s under control.

So now they’re asking you to look directly into a camera so a machine can confirm you’re real.

This is the part of the AI boom nobody talks about at venture pitch meetings. The infrastructure of trust—the basic assumption that the person you’re talking to or the account you’re following actually exists—is collapsing. And the solution isn’t better moderation or smarter algorithms. It’s biometric verification. It’s forcing humans to prove they’re human in their own digital spaces.

My read is this is going to accelerate dramatically. Not because it’s elegant or fair, but because the alternative—a completely unreliable information environment—is worse.

Photo by nappy / Pexels

The IPO Wave Is the Real Story

While everyone’s debating whether Mythos poses an existential threat, Cerebras is filing to go public. SpaceX. Anthropic. OpenAI. The wave of enormous initial public offerings is happening right now, and it’s basically a bet that all this works out. That society figures out how to manage these tools. That regulation doesn’t kill the upside. That the money keeps flowing.

This is the moment where venture capital stops being patient and starts demanding returns. It’s where the experimental phase ends and the business phase begins. And that changes everything about incentives.

Here’s what I think gets lost in the coverage: the IPO wave isn’t about confidence in AI. It’s about confidence in the market’s ability to absorb and price AI. The companies going public aren’t betting the world will be fine with superintelligent systems. They’re betting investors will be fine with it, which is a totally different calculation.

When a chip maker like Cerebras files its prospectus, it’s not because they’ve solved the safety questions. It’s because demand is so strong they can’t build chips fast enough. And when there’s that much money on the table, regulatory friction becomes a feature, not a bug—it keeps new competitors out.

The Measurement Obsession Says Everything

METR—a nonprofit AI organization—created a chart that measures the development of big AI systems, and the entire industry became obsessed with it. This is the kind of detail that sounds boring until you realize what it means: tech’s favorite game is quantification. If you can measure it, you can game it. If you can game it, you can monetize it.

The chart probably measures something real. But the obsession with it reveals something darker. It’s the same instinct that made everyone stare at burn rates in 2000. The same energy that made people refresh Substack stats at midnight. We’re all looking for a single number that tells us if we’re winning.

Spoiler: there is no such number.

But right now, everyone’s watching METR’s chart like it’s the S&P 500, which means everyone’s optimizing for what shows up on the chart rather than what actually matters. That’s how you get systems that look spectacular on metrics and fail catastrophically in the real world.

The “Digital Twin” Question Is Actually a Legal Nightmare

Could a digital twin make you a superworker? Could it? Sure. Will it? Depends entirely on whether it’s legal.

And nobody knows if it’s legal.

The question sounds like a fun productivity story—imagine: a digital version of you that works while you sleep. But it’s actually a legal minefield. Who owns the digital twin? If it makes mistakes, who’s liable? If it violates copyright by processing training data, is that your liability or the company’s? If it replicates your decision-making but differently, have you been cloned?

This is where the AI boom meets reality. You can build it. You can sell it. You can promise it’ll make everyone 40% more productive. But the moment someone sues, the entire thing might collapse because we don’t have legal frameworks for this yet.

My prediction: we’ll see the first major digital-twin lawsuit by Q4 2025. It’ll be messy. It’ll probably involve employment law and IP law and something totally unforeseen. And the companies pushing digital twins the hardest will suddenly become very quiet about rollout timelines.

Photo by Markus Spiske / Pexels

The UK’s Move Is the Template Everyone Will Copy

Keir Starmer told tech bosses that things can’t go on like this with online safety, and the UK is seriously considering banning under-16s from social media. This sounds drastic until you remember that regulation in the UK often becomes regulation everywhere—the GDPR proved that one country’s rules can reshape the entire global internet.

If the UK bans under-16s from social media, expect the EU to follow. Expect South Korea to follow. Expect California to try something similar and fail, then try again with a different angle.

The age-ban approach is crude. It probably won’t work perfectly. But it has a political virtue that matters: it’s simple. It’s testable. It doesn’t require tech companies to solve AI moderation. It just says: no kids.

Is this the answer? No. But it’s an answer, and right now, governments are desperate for anything that looks like action. The companies will fight it. They’ll lose some battles and win others. But the velocity has shifted. We’re not in the “tech companies police themselves” era anymore.

The Undeniable Pattern

Here’s what I’m genuinely uncertain about: whether the government’s “productive” meeting with Anthropic actually means anything. Did they negotiate restrictions on Mythos deployment? Did they agree to share research? Did they just agree to keep talking?

I don’t know. None of us do.

But here’s what I’m sure of: the era where AI firms could move fast and break things is ending. Not because the tech stopped being powerful. Because the things it can break got too important to ignore.

The iris scans. The fake avatars. The digital twins. The IPO wave. The regulatory conversations happening in rooms we can’t see. It’s all the same story. The infrastructure of trust is fracturing. The legal questions have moved from theoretical to urgent. And money is flooding toward the exits before everything gets more complicated.

What I’m Watching

• Anthropic’s first earnings call post-IPO (if it happens by Q3 2025): Watch how they discuss Mythos deployment constraints. If they mention government restrictions, that’s your sign the White House meeting had teeth.

• The first digital-twin employment lawsuit: Specifically, whether it involves IP infringement or liability for the twin’s decisions. This will set the template for everything else.

• UK Under-16 Social Media Ban enforcement timeline: If they actually implement it by late 2025, watch whether the enforcement is technical (biometric age verification—circles back to Tinder/Zoom’s iris scans) or legal. This tells you how hard they’re willing to push.

• METR chart divergence from actual AI capability: When a system scores high on their metrics but fails spectacularly in the real world, that’s your signal the measurement obsession has hit its limit.

The reckoning isn’t coming. It’s here. We’re just still deciding what it means.