The Year AI Got Too Dangerous to Ignore (And Too Useful to Ban)

---

The White House called a meeting with Anthropic last Friday. Not a photo op. Not a listening session. An actual negotiation.

This matters because it signals something that’s been building for months: we’ve crossed into a new phase of AI where the technology is too strategically valuable for governments to regulate aggressively, but too risky for them to ignore. That tension is about to reshape everything from how we hire people to what happens to social media.

When the Government Needs You More Than You Need It

Anthropic’s Mythos model does something no previous AI has done in public: it outperforms humans at hacking and cybersecurity tasks. That’s not theoretical. That’s not marketing copy. U.S. officials believe this technology could be “critical for security.” So critical that instead of threatening to regulate it out of existence, they’re trying to work out a deal.

This inverts the usual script. For a decade, tech companies complained about overregulation while governments scrambled to catch up. Now the government is the one trying to stay relevant. The White House meeting after Mythos launched wasn’t about restriction—it was about access and terms. Anthropic’s leverage just went through the roof.

Compare this to 2022, when regulators were talking about AI as purely a consumer protection issue. Back then, the risk was deepfakes and chatbot bias. Now it’s: can we keep this technology from making critical infrastructure trivial to break into? The conversation shifted from “should we slow this down?” to “who gets to use this and under what conditions?”

My read: this is the beginning of a two-tier AI system. One version of these models will exist in a tightly controlled space where governments and vetted corporations can access Mythos-level capabilities. Everything else gets the consumer-facing version with guardrails bolted on. Anthropic just became a de facto part of national security infrastructure.

Photo by Eduardo Romero / Pexels

The Iris Scanner Arms Race

Here’s where the absurdity kicks in: while Mythos is so powerful governments worry about it, TikTok and Zoom are asking you to scan your eyeball to prove you’re real.

This isn’t paranoia. Hundreds of fake AI-generated avatars have swarmed platforms like TikTok and Instagram specifically to target conservative voters. The deepfakes are getting good enough that basic photo verification doesn’t work. So the solution is biometric: iris scans. Zoom and Tinder are now asking for eye-scans to validate accounts.

There’s something almost darkly funny about this escalation. We built AI so good that photographs aren’t evidence anymore. Now we need to scan your iris. Next year? DNA? The problem compounds faster than the solution ever could.

But here’s what actually matters: biometric verification is the wedge that opens the door to a surveillance infrastructure most democracies have resisted. Once iris-scanning becomes normalized for social media access, the expectation spreads. Banks want it. Governments want it. And suddenly the infrastructure for mass identification—the kind of thing civil libertarians have fought for decades—arrives not through legislation but through “combating AI-generated fakes.”

I think this will become the dominant narrative that justifies biometric systems we should be more skeptical of. We’re trading one problem (fake accounts) for a much larger one (comprehensive digital identity tracking). And it’ll happen so gradually, so sensibly, that most people won’t notice until it’s standard.

The Superworker Problem Nobody’s Ready For

Buried in the week’s headlines is something quieter but potentially more destabilizing: digital twins are becoming real.

Companies are now selling “digital twins”—AI versions of your work self that handle your tasks while you’re not there. It’s being pitched as the solution to productivity. Firms say employees become “superworkers” when they’ve got an AI clone handling routine work. It sounds great until you think about what it means for hiring, wages, and job security.

Here’s the legal minefield nobody wants to talk about: if your employer has an AI version of you, do they own it? Can they use it after you leave? If the digital twin performs better than you, what happens to your employment? These questions don’t have answers yet, which means someone’s going to have a very expensive court case before the law clarifies.

But the real issue is economic. Digital twins are the automation play that actually works—not robots replacing factory workers, but AI replacing knowledge workers. And it’s happening right now, in spreadsheets and customer service queues, before regulation even catches up.

Photo by nappy / Pexels

When London Tells Silicon Valley “No”

The UK government is making moves that should terrify tech companies if they were paying attention.

UK Prime Minister Starmer told tech bosses this week that “things can’t go on like this with online safety.” The government’s actively consulting on banning under-16s from social media entirely. This isn’t a fringe proposal. This is a major Western government genuinely considering outright prohibition.

This is different from regulation. Regulation is about rules and compliance. Prohibition is about saying: this platform is not allowed to exist for this age group. Period.

If the UK actually bans under-16s from social media, the EU and probably Australia will follow within 18 months. That’s roughly 15% of TikTok’s global user base eliminated overnight. Snap, Instagram, YouTube—all face the same cut. The financial impact would be brutal.

What’s weirdest is that tech companies seem unprepared for this actually happening. They’ve lobbied against it. They’ve promised to do better. But they haven’t fundamentally changed what their products do—addict young people with algorithmic feeds. So government is moving toward the nuclear option: make the addiction illegal.

I genuinely don’t know if this will work or just create a black market for TikTok. But I’m confident it’s coming.

The IPO Wave That Changes Everything

Cerebras just filed to go public. SpaceX, Anthropic, and OpenAI are all preparing public offerings. We’re about to see an IPO wave that’ll push billions into AI infrastructure and companies.

This matters because public markets have patience for moonshots that venture capital is getting tired of. We’re about to find out which AI bets actually have sustainable business models and which ones were just riding hype. Some of these companies will crater on the first quarterly earnings call when investors realize “exciting technology” doesn’t equal “profitable company.”

But more importantly: once these companies are public, they can’t pivot on a dime. Anthropic’s government relationship will be documented in SEC filings. OpenAI’s charitable structure becomes a public scandal. SpaceX’s Starlink gets used for things the founders didn’t plan on. The private bet becomes a public commitment.

My prediction: within two years, at least one of these IPOs will be revealed to have business model problems that crash the stock 60%. And it’ll trigger a broader conversation about whether AI companies actually have defensible economics or if we’ve been funding a bigger bubble than 2000.

What I’m Watching

• Anthropic’s public filing (expected Q2 2025): Watch for how much revenue comes from government contracts vs. consumer/enterprise. If government deals are >40% of revenue, we’re officially in a two-tier system.

• UK ban legislation passage timeline: If the under-16 social media ban passes into law before mid-2025, expect copycat legislation in Australia, Canada, and EU countries within 90 days. This cascades fast.

• Cerebras and AI chip maker profitability: These companies will report Q1 2025 earnings. If any are still unprofitable on a unit basis after scaling production, the entire semiconductor AI bet gets questioned. Watch gross margins specifically.

• The first iris-scan biometric security breach: It will happen. When it does, it’ll create political pressure to regulate biometric data differently than passwords. This is where the narrative around “necessary security” gets challenged.