The AI Safety Theater Is Over. Now the Real Mess Begins.

---

Anthropic built a model so dangerous they won’t let anyone use it.

Then they decided who gets access anyway.

That’s not a safety guardrail. That’s security theater, and we’re watching it collapse in real time. The Mythos AI model—which Anthropic explicitly says has hacking capabilities too risky for public release—has already triggered emergency responses from central banks and intelligence agencies. The company’s solution? Become the gatekeeper. Stay in control. Decide which governments, corporations, and researchers deserve to play with the dangerous toy.

This is the moment when “responsible AI” stops being a philosophy and becomes a business model. And it’s not going well.

Photo by Inaamulhaq . / Pexels

When “Dangerous” Becomes a Feature, Not a Bug

Let’s be precise about what’s happening here. Anthropic didn’t accidentally create a model with hacking capabilities. They built it that way. The question—the only question that matters now—is whether limiting access actually stops bad actors, or whether it just slows them down long enough for Anthropic to sell the narrative of responsible stewardship.

I think it’s the latter.

Here’s why: The moment you admit a model can hack, you’ve already lost control of it. You’ve just hidden the loss. Intelligence agencies and nation-states aren’t waiting for Anthropic’s permission. They’re either building their own dangerous models or finding ways to access Mythos through back channels. Anthropic’s access control becomes theater for regulators and the public—proof that someone is thinking about safety while the real players move in the shadows.

Meta, meanwhile, is playing an entirely different hand. They’re not even pretending anymore.

The Panopticon Goes to Work

Meta announced it will track employees’ clicks and keystrokes to train AI models. Not for security. Not for compliance. For training data. They’re instrumentalizing their own workforce—turning every keystroke, every decision, every mistake into fodder for machine learning systems.

This isn’t new in principle. Companies have monitored workers for decades. But there’s something particularly blunt about doing it openly, at scale, to build AI. It’s like announcing you’re going to harvest organs from your employees to research transplants. Technically you own the workplace. Ethically? You’ve just admitted your priorities.

What gets me is the precedent. Once Meta does this and faces minimal consequences (a few op-eds, maybe a regulator stern email), every other tech company will follow. Google. Amazon. Microsoft. Within 18 months, treating employee behavior as raw AI training material will be standard practice. We won’t even debate it anymore.

The Copyright Apocalypse Is Actually Here

While everyone’s arguing about whether AI training on copyrighted work is legal, the practical answer has already arrived: it doesn’t matter.

Leaked code from Anthropic’s Claude shows how fast these tools can now reproduce creative work. Faster than ever. The question isn’t whether it’s technically possible—it’s been possible. The question is whether copyright law can even exist in a world where a teenager with $20 and a laptop can generate commercially indistinguishable artwork in seconds.

My prediction: Copyright as we know it dies by 2027. Not legally—courts will keep ruling on edge cases forever. But practically. Artists and writers will eventually accept that control over their work is gone, and the industry will reorganize around whoever can generate the most compelling outputs the fastest, not whoever created the original. The legal system will limp along behind, writing rules for a world that no longer exists.

Photo by nappy / Pexels

Why Peter Molyneux Matters More Than You Think

Peter Molyneux said he’s making one last game—Masters of Albion—then he’s done. The creator of Fable and Black & White, one of the most inventive minds in gaming, is retiring from game-making at the moment when AI could theoretically let him create worlds faster and cheaper than ever before.

He’s not excited about that possibility. He’s leaving.

That tells me something. The industry’s best creative minds aren’t optimized for the AI future. They built their careers on a scarce resource: human imagination, carefully crafted over years. AI doesn’t reward that kind of depth. It rewards scale, speed, iteration. It’s not a tool for Peter Molyneux. It’s a replacement for him.

Microsoft knows this. They’re cutting Game Pass prices but pushing back when new Call of Duty games hit the service—“about a year” after release instead of day-one. Translation: Game Pass was becoming too attractive to players relative to game sales. They need to rebuild the incentive structure. The model was unsustainable. The price cut is actually a retreat.

Gaming is entering a new era where the economics are totally different. Indies can’t compete. AAA studios face margin pressure. And subscription services—the thing that was supposed to save gaming—are actually fragmenting the market even further. Everyone’s optimizing locally, and nobody’s building for a coherent future.

The Crypto Cannibalism

Justin Sun suing the Trump family’s crypto venture—World Liberty Financial—for fraud. This is what crypto’s mainstream moment looks like. Not innovation. Not revolution. Just rich people fighting over who scammed whom.

The fact that Sun was a backer and then sued suggests he got burned or realized the whole thing was hollow. Either way, it’s the clearest possible sign that crypto’s institutional moment has passed. No technology moves from “fringe” to “fraud litigation with celebrity families” without losing whatever made it interesting.

SpaceX’s Strange New Direction

Elon Musk is shifting SpaceX’s goals away from Mars and toward “moonshots” as the company prepares to go public. That’s strategic language for “we’re not actually going to Mars, please buy our stock.” The Cursor deal for $60 billion suggests SpaceX is now explicitly AI-focused rather than space-focused.

This is fine. Maybe it makes business sense. But it’s worth noting: the company that spent 15 years building rockets to reach another planet is now pivoting to whoever’s paying the most and positioning themselves as an AI play for Wall Street.

Photo by Markus Spiske / Pexels

Here’s What Actually Worries Me

I can handle most of this. Dangerous AI models? Intelligence agencies will probably manage it. Meta’s surveillance? Regulations will eventually catch up (slowly, badly, but eventually). Copyright collapse? Artists will adapt. Crypto fraud? Market will correct.

What I genuinely don’t know is whether the institutions meant to oversee all this have any actual power left. Ofcom is investigating Telegram for child safety failures, and Telegram denies everything. What happens next? Probably a fine in 18 months. Probably nothing changes. Probably Telegram knows they’re not actually accountable to regulators in any meaningful way because they’re too distributed, too international, too useful.

That’s the pattern now. Admit nothing. Deny everything. Wait it out. Institutions move slowly. Tech moves fast. The gap between them widens every year.

What I’m Watching

• Anthropic access logs. If hackers manage to access Mythos within the next 6 months, the entire “responsible gatekeeper” model collapses. Watch for any zero-day exploits or leaked Mythos instances. That’s the real test of whether access control actually works or just creates a false sense of security.

• Meta’s employee attrition rate. If they lose top talent because of the keystroke tracking program, that’s a signal that even internally, there’s pushback. If nothing happens and employee flight stays normal, then we know this precedent will stick industry-wide.

• Copyright litigation outcomes in 2025. Specifically, watch for the first major case where an AI company wins the right to train on copyrighted work without permission. That verdict will reshape the entire creative industry’s economics overnight.

• SpaceX’s IPO valuation. If they’re valued primarily on AI potential rather than launch capacity, that’s confirmation that even hardware companies now need to sell themselves as AI plays to justify prices. That means the bubble’s getting rounder.