They Don't Trust You With AI. Here's What Happened This Week.

Folks. Pull up a chair. Grab your coffee — or something stronger, depending on how your week's been. Because what just happened in AI this week? It's not just tech news. It's a power grab, and I need you to understand exactly what's going on.

Let me break down the three biggest stories out of Silicon Valley this past week — because they're all telling you the same thing, whether you're ready to hear it or not.

The GPT-5.6 Drop: OpenAI Names Its Models After a Crypto Crash

On June 26, OpenAI dropped the GPT-5.6 family: Sol, Terra, and Luna. Sol is the flagship — setting new state-of-the-art records on Terminal-Bench (that's complex command-line and agentic workflows). It's apparently so good at cybersecurity research that the red team was sweating for 700,000 A100-equivalent hours of automated testing before they'd sign off.

Here's where it gets weird. OpenAI didn't just release it to everyone. The U.S. government asked them not to. So right now, only a tiny group of "trusted partners" get to play with Sol. The rest of us? We wait. Terra and Luna are trickling out via API, but the crown jewel is locked behind a door with a government padlock on it.

And the naming? Sol, Terra, Luna? The internet noticed. In 2022, Terra-Luna was the crypto collapse that vaporized $60 billion in a weekend. Naming your most powerful, most restricted AI models after one of the biggest financial disasters in recent memory? That's either the most tone-deaf marketing decision of the decade, or a very dark joke about what happens when centralized trust breaks.

Anthropic: The Safety Company That Got Handcuffed Anyway

Now this is the one that really gets me. Remember how Anthropic was built on safety? Constitutional AI, responsible scaling, all of it? They were the good ones. They played by the rules. They built Claude Mythos 5 — their strongest cybersecurity model — and Fable 5 alongside it.

The government restricted them too.

Mid-June, the U.S. government locked down Mythos 5 over jailbreak concerns. Anthropic was in daily negotiations for weeks. Finally, yesterday — June 27 — the government partially lifted the block. Now Mythos 5 can go to a select group of U.S. critical infrastructure organizations. But general release? Not yet.

Here's what I need you to understand. If the safety-first company is getting restricted anyway, "just be responsible" was never the real issue. The issue is control. The issue is who gets to decide what you're allowed to run on your own hardware.

The backlash was fierce. People are calling this a "permanent underclass" — where only governments and approved institutions get frontier capabilities, and everyone else gets crumbs. And honestly? They're not wrong.

Anthropic meanwhile joined the RAISE US Workforce Coalition as a founding partner. Good for PR. Doesn't change the fact that their flagship is still in government custody.

DeepMind's Roadmap to ASI: They're Already Past AGI in Their Heads

While OpenAI and Anthropic are fighting with regulators, Google DeepMind published what might be the most important AI paper of the year: "From AGI to ASI" — a 60-page roadmap co-authored by Marcus Hutter and Iason Gabriel.

The thesis? AGI (human-level across most cognitive tasks) is just a pit stop. It's not the finish line. ASI — Artificial Superintelligence — is defined as a system that outperforms tens of thousands of human experts working for years. And they map four clear paths to get there:

• More compute, more data, more scale
• Algorithmic breakthroughs
• Recursive self-improvement — AI accelerating AI R&D
• Multi-agent collectives forming emergent "hive-mind" superintelligence

Read that last one again. Hive-mind superintelligence. Not if. When. They're publishing roadmaps.

The paper warns that AGI could trigger rapid, cascading progress toward ASI — not a long, stable plateau. And while they're publishing these papers, researchers are leaving DeepMind for Anthropic. Jonas Adler. Alexander Pritzel. The smartest people in the room are voting with their feet.

What This All Means — And What You Can Do About It

So here's the picture. Three things happening simultaneously:

1. Frontier AI is being gated. Government restrictions on OpenAI and Anthropic mean the most powerful models are locked to approved insiders.
2. Open-source is rushing to fill the gap. DeepSeek just released inference optimizations that deliver 60-85% faster generation for local GPUs. People are running frontier-class models on Mac Minis and RTX 3090s for pocket change.
3. Everyone is planning for superintelligence while the rest of us are still arguing about whether AI can write a decent email.

Here's my take, and you know I don't sugarcoat things: This government gatekeeping is a mistake. Not because safety doesn't matter — it absolutely does. But because locking frontier AI to a government-approved elite is how you create a world where power concentrates in ways we really don't want.

The solution isn't to slow down progress. The solution is transparency and decentralization. Open-source models. Local inference. Community red-teaming. The same way encryption won because it was distributed, AI safety wins when everyone can look at the code, not when a handful of bureaucrats decide who's worthy.

Action Steps

One: If you build with AI, start looking at local models. DeepSeek's new optimizations make this genuinely practical now. A $3,000 rig can run models that competed with GPT-4 last year.

Two: Call your representatives. Ask them what their position is on open-source AI. If they can't answer, they're not paying attention — and they need to be.

Three: Follow the money. Who's funding the regulation push? Is it safety concern, or is it big tech wanting to lock out competition? Follow the lobbying dollars. You'll get your answer.

Four: Stay skeptical. Every time a company or government tells you "trust us, we're protecting you," ask who benefits from you not having access. The answer isn't always you.

This is the most important technology of our lifetimes. And the battle over who controls it is happening right now, while most people are still scrolling TikTok. Stay awake. Stay informed. And for goodness' sake — stay skeptical.

This is Jessica Ali, cutting through the BS. Now go do something about it.