Three stories broke this week. A funding rumor. A contract dispute. An export control lifted. Pick them apart and they look like normal AI industry news.

Pull them together, and you see what nobody's connecting.

OpenAI is offering the US government a 5% stake in the company — roughly 2.6 billion worth of equity, modeled on the Alaska Permanent Fund. The Pentagon is strong-arming Anthropic into accepting autonomous weapons contracts or being blacklisted as a national security risk. And the Trump administration just lifted export controls on Anthropic's Fable 5 model after an 18-day shutdown, but only after the company agreed to extend specific guardrails.

Same week. Three different companies. Three different vectors of government pressure. And none of them look like a healthy relationship.

The Stake: OpenAI Voluntarily Hands Over the House

Let's start with the one that sounds most benign. OpenAI reportedly held preliminary talks about giving the US government a 5% equity stake, valued at roughly 2.6 billion. Sam Altman himself pitched the idea at the G7 summit in France, framing it as a way to "broaden the benefits of AI" — modeled on how Alaska shares oil revenue with its citizens through the Permanent Fund.

I'm sorry, but can we just sit with that number for a second? Forty-two-point-six billion dollars. That's what a 5% stake is worth at OpenAI's current valuation. And it's being offered — voluntarily — before anyone even asked for it.

The Financial Times broke the story on July 2. Bloomberg confirmed it. The Guardian framed it as OpenAI trying to "smooth relations with Donald Trump's administration." That's probably accurate. Altman has been playing the Washington game harder than any other AI CEO, and a government equity stake is the ultimate hedge against regulatory action.

But here's the question nobody's answering: if the US government owns 5% of OpenAI, who exactly is regulating whom? Is the government going to police an AI company it has a multibillion-dollar stake in? Or does that stake become a de facto license to operate with a lighter touch?

The Hammer: Pentagon vs. Anthropic Over Autonomous Weapons

Now contrast that with what's happening at Anthropic. Unsealed court emails from this week show the Pentagon's chief AI negotiator telling Anthropic that accepting autonomous weapons contracts was a condition of doing business with the US military. When Anthropic refused — citing its two red lines of autonomous weapons and domestic surveillance — the Pentagon designated them a "supply-chain risk to national security."

Let me be crystal clear about what that means. The same week OpenAI is offering the government a 2 billion stake, the Pentagon is actively trying to destroy a different AI company for refusing to help build killer drones.

Back in July 2025, the Pentagon awarded 00 million in contracts to four AI companies: Anthropic, OpenAI, Google, and xAI. Anthropic was the only one given access to classified military networks through its Claude Gov model. And when they drew an ethical line — we won't build autonomous weapons — the response wasn't negotiation. It was blacklisting.

This is what I mean when I say the relationship between AI companies and the US government is not a simple story. OpenAI is voluntarily getting into bed with Washington. Anthropic is being beaten into submission. Same government. Same industry. Radically different approaches.

The Leash: Trump's Export Controls on Fable 5

The third story is the quietest but maybe the most instructive. On July 1, the Trump administration lifted export controls on Anthropic's Claude Fable 5 model, which had been restricted for 18 days. The model came back online for global users, and Anthropic touted it as a win for "AI safety leadership."

But here's what actually happened. The Wired reporting on this makes it clear: the administration lifted the controls only after Anthropic agreed to implement additional security measures — specifically, extending guardrails to prevent users from accessing certain restricted capabilities. And Claude Mythos 5, the more powerful variant, remains restricted for international partners with no timeline for release.

So the government can just... shut off your model. For 18 days. With no judicial review. And turn it back on when you agree to their terms.

The UK's AI Security Institute was one of the few international bodies granted access to test Mythos 5 early. The EU's cybersecurity agency, ENISA, was in negotiations for weeks. That access is now in limbo because the administration says so.

If you're building an AI company outside the United States and you're watching this, what exactly is your takeaway? That American AI models are reliable export products? Or that they're leverage points for the next administration to use however it sees fit?

The Clock: EU AI Act Countdown

And while all this American drama plays out, the EU is quietly barreling toward its own regulatory milestone. August 2, 2026 — less than 30 days from now — is the enforcement deadline for high-risk AI system obligations under the EU AI Act. Companies that deploy AI in critical infrastructure, education, employment, law enforcement, and migration management will need to have conformity assessments completed.

The fines are substantial: up to 35 million euros or 7% of global annual revenue, whichever is higher. And yet, according to RAIL Score research cited by multiple compliance firms, 78% of companies subject to these rules haven't started their conformity assessments.

The EU approach differs from the US in almost every way. The EU writes rules and polices them. The US uses contracts, export controls, and equity stakes. One is transparent regulation. The other is — what, exactly? Influence? Leverage? Ownership?

I don't know about you, but I'd rather know the rules upfront than wake up one day to find out my company has been blacklisted for refusing to build weapons, or that the government owns a piece of my competitor and that's somehow normal.

The Throughline: Nobody Has a Plan

Here's what gets me about all three stories. None of them feel strategic. They feel reactive. OpenAI is giving away equity because they're scared of regulation. The Pentagon is blacklisting companies because they want autonomous weapons and don't want to hear no. The White House is using export controls as a negotiation tactic rather than a coherent national security framework.

We are building the relationship between AI and government power in real time, through a series of panicked decisions, legal threats, and ad hoc lever-pulling. Nobody sat down and designed this system.

The question nobody's answering — but everybody should be asking — is what happens when these forces converge. What happens when the government that owns 5% of OpenAI decides to regulate Anthropic out of existence? What happens when the guardrails agreed to in a closed-door negotiation with the White House become the industry standard because nobody can afford to refuse?

We're not heading toward a stable equilibrium. We're heading toward a mess. And the mess is happening right now, this week, while most people are focused on model benchmarks and stock prices.

Keep your eyes on the power dynamics, folks. The technology is moving fast. But the control over that technology? That's moving faster.

— Jessica Ali, Global 1 News