The $25 Billion Bet That Backfired? Why Amazon’s CEO Just Helped Shut Down Its Own AI Investment

Andy Jassy sounded the alarm on Anthropic’s AI models, but the economics reveal a surprising twist

---

You know that sinking feeling when you lend your car to a friend, and then you have to call the cops because he’s doing something reckless with it?

That’s roughly what happened in the AI world last week.

Andy Jassy, CEO of Amazon, the largest investor in the AI startup Anthropic, walked into meetings with senior Trump administration officials and raised concerns about security risks in Anthropic’s most advanced AI models.

Days later, the U.S. government issued an export control directive that forced Anthropic to shut down global access to those very models.

Wait, what?

Why would the CEO of a company that just pumped $25 billion into an AI startup turn around and help the government disable its flagship products?

I’ll be honest: my first reaction was exactly that. Betrayal. Self‑sabotage. A corporate drama worthy of a Netflix series.

But after digging into the details, I realized the story is far more interesting, and far less dramatic, than the headlines suggest.

Let me walk you through what actually happened, why it makes perverse sense, and what it means for the future of AI regulation.

---

From Closed‑Door Meeting to Global Shutdown

Here’s the timeline.

On June 9, Anthropic publicly launched Claude Fable 5 and Mythos 5, its most powerful AI models yet. Fable 5 was the “safer” version, wrapped in extra guardrails. Mythos 5 was the raw, unbridled powerhouse, released only to trusted partners.

Three days later, the Commerce Department issued an unprecedented directive: all foreign nationals, including Anthropic’s own foreign‑born employees, were banned from accessing either model.

The trigger?

A series of private conversations between Andy Jassy and officials including Treasury Secretary Scott Bessent.

Jassy told them that Amazon researchers had used Anthropic’s Fable 5 to obtain information that could be used in cyberattacks, information that was supposed to be off‑limits.

Let that sink in.

Amazon, Anthropic’s biggest financial backer, was also the one doing the security testing that exposed its vulnerabilities.

The company that writes the biggest checks was also the one holding the metaphorical lock‑picking kit.

---

The “Jailbreak” That Broke the Camel’s Back

So what exactly did those Amazon researchers find?

In cybersecurity terms, they discovered a jailbreak, a method to bypass the safety guardrails built into Fable 5.

Jailbreaking isn’t inherently nefarious. Security researchers do it all the time to stress‑test systems. Think of it like a locksmith picking a lock to prove it can be picked.

In this case, Amazon’s team used a series of carefully crafted prompts to get Fable 5 to identify a small number of software vulnerabilities, some of which were already known and relatively minor.

Anthropic pushed back hard. In a public statement, the company said: “These vulnerabilities all appear relatively simple, and we have found that other publicly‑available models are able to discover them as well without requiring a bypass.”

They also noted that no tester, including those from the U.S. government and the UK’s AI Safety Institute, had found a universal jailbreak (a method that broadly bypasses all safeguards).

But the government didn’t wait for a universal threat. The existence of any reliable jailbreak was enough to trigger an export control action.

Anthropic was given the directive at 5:21 pm ET on a Friday. No specific details. No time to negotiate. Just a blunt order: block access or else.

The result? Anthropic had to abruptly disable both models for all customers to ensure compliance, because the restriction applied even to foreign nationals inside the U.S.

---

The $25 Billion Paradox

Now here’s where the story gets genuinely strange.

Two months before the shutdown, Amazon had announced an expanded agreement with Anthropic: up to $25 billion in new investment, bringing Amazon’s total potential stake to $33 billion, and a commitment from Anthropic to spend over $100 billion on AWS over the next decade.

At Anthropic’s latest valuation, $965 billion after a $65 billion Series H funding round in May, Amazon’s stake is worth roughly $135 billion.

So again: why would Jassy poke the bear? Why would the CEO of a company with that much financial exposure voluntarily raise red flags about the very models his investment depends on?

The easy answer is national security. Jassy may have genuinely believed the risks warranted government action.

The more interesting answer, and the one that explains the economic logic, is that AWS profits whether Anthropic succeeds or fails.

---

The “Landlord” Economics

Let me explain with a simple analogy.

Imagine you’re a landlord. You lease office space to a high‑growth tech startup. The startup pays you a massive, multi‑year lease upfront.

Now imagine that startup runs into regulatory trouble. Maybe its product gets banned in certain markets. Does that affect your rent?

No.

You already collected the lease payments. The startup’s problems are its own. You still own the building.

That’s exactly how Amazon structured its deal with Anthropic.

The $25 billion investment isn’t just an equity bet, it’s a down payment that locks in a massive cloud customer. Anthropic committed to spending $100 billion on AWS over ten years. That revenue is contracted, regardless of whether Fable 5 or Mythos 5 ever sees the light of day again.

As one analyst put it, “AWS makes money from that $100 billion whether Anthropic’s models are banned or not. The cloud infrastructure business sells computing power. It doesn’t matter if the tenant inside the building gets evicted, the landlord already collected the lease.”

This is a structural shift that most coverage missed.

Infrastructure providers like AWS profit from model companies’ regulatory risks. When a model gets banned, the compute contract doesn’t disappear. If anything, it tightens Amazon’s grip on Anthropic’s infrastructure future.

So Jassy raising concerns with Trump officials isn’t necessarily an act of war against his own investment. It’s an act of risk management, one that leaves Amazon’s core cloud business untouched while potentially steering regulatory outcomes.

---

The Government‑Enforced Moat

Here’s the counterintuitive twist that might make your head spin.

Regulation isn’t always a penalty.

Sometimes, it’s a barrier to entry.

Think about it: if the U.S. government blocks foreign access to Anthropic’s most advanced models, that doesn’t hurt Anthropic’s domestic business. It actually strengthens Anthropic’s position relative to foreign competitors who can’t access those models.

In fact, the shutdown can be read as implicit government validation of Anthropic’s technical lead. If the models weren’t genuinely powerful, if they weren’t worth restricting, why would the government bother?

This is what industry watchers call a government‑enforced moat.

The same regulatory action that looks like a blow to Anthropic’s global ambitions also reinforces its value for U.S.‑based customers and partners.

And for Amazon, that moat means Anthropic becomes even more dependent on AWS as its primary infrastructure provider, because if Anthropic wants to rebuild access or launch new models, it needs compute. And who controls that compute? AWS.

---

The Wider Regulatory Storm

The U.S. crackdown isn’t happening in a vacuum.

Around the world, regulators are circling Amazon’s relationship with Anthropic.

• In the UK, the Competition and Markets Authority (CMA) launched an inquiry into the partnership, questioning whether Amazon’s minority stake and exclusivity arrangements constitute a backdoor merger.
• In Brazil, antitrust authority CADE delayed a ruling on its own investigation, saying it needed more time to assess whether Amazon’s deal should have been notified for merger review.
• The DOJ and FTC have also been probing cloud‑AI alliances across the board, treating generative AI as subject to consumer‑protection and antitrust law.

And that’s just the competition angle.

On the national security front, President Trump signed an executive order in early June requiring AI companies to grant the government voluntary access to new models before public release, a framework clearly aimed at Anthropic’s Fable 5 rollout.

Then there’s the ongoing Pentagon dispute. In January 2026, Anthropic CEO Dario Amodei refused to remove safety guardrails that would have allowed surveillance applications. The Pentagon responded by designating Anthropic a supply chain risk and killing a $200 million contract.

Anthropic is now suing the administration over that designation, but the damage is done. The company has been publicly painted as a security concern from multiple angles.

---

Where Do We Go From Here?

So what does this all mean for the future of AI regulation?

Three scenarios are worth watching:

1. Temporary pause. If the jailbreak concerns are resolved and Anthropic demonstrates stronger safeguards, the government could lift the export controls. In this scenario, Anthropic returns to the global race and Amazon’s investment thesis stays intact.

2. Permanent restriction. If the U.S. decides that frontier AI models are simply too dangerous for broad foreign access, export controls become a permanent feature of the landscape. That would tilt the competitive playing field sharply in favor of U.S.‑based companies and their infrastructure partners.

3. Selective re‑opening. Perhaps the most likely outcome: the government allows access on a case‑by‑case basis, with strict licensing requirements for foreign customers. This creates a bureaucratic moat that benefits incumbents like Anthropic and their cloud partners.

For Amazon, none of these scenarios is catastrophic.

The $100 billion cloud commitment is already locked in. The government’s action doesn’t unwind that contract, it just changes who can use the models running on top of it.

For Anthropic, the picture is more complicated. The company is still pursuing a $1 trillion IPO. It just raised $65 billion at a $965 billion valuation. But being the first AI startup to have its flagship models disabled by government order is not the kind of headline you want in your S‑1 filing.

Still, Anthropic may emerge from this stronger. Its safety‑first brand, once seen as a marketing gimmick, now looks prescient. In a world of tightening compliance, the company that already invested in guardrails may be better positioned than rivals who raced ahead without them.

---

The Takeaway

The story of Andy Jassy’s talks with U.S. officials isn’t a simple tale of corporate betrayal.

It’s a window into a new kind of economic reality: infrastructure wins.

Cloud providers like AWS make money on compute, regardless of which models succeed or fail. Model builders like Anthropic are subject to regulatory whiplash, but their deepest partners are insulated by long‑term contracts and diversified exposure.

And government action, which looks like a hammer coming down, can actually function as a competitive shield, raising barriers for everyone else while locking in incumbents.

So the next time you see a headline about a tech CEO “betraying” their own investment, take a second look.

Follow the money. Look at the contracts. Ask who’s truly exposed, and who’s just collecting rent.

You might find that the story isn’t about betrayal at all.

It’s about who owns the building.

---

Key Takeaways (Quick Reference):

• Amazon CEO Andy Jassy raised security concerns about Anthropic’s AI models in meetings with Trump administration officials, triggering a U.S. export control ban on Fable 5 and Mythos 5.
• The ban forces Anthropic to block all foreign nationals from accessing the models, including its own foreign‑born employees.
• Amazon is Anthropic’s largest investor, with a potential $135 billion stake and a $100 billion cloud spending commitment from Anthropic.
• Despite the investment, Amazon’s AWS revenue from the cloud contract is secured regardless of regulatory outcomes, making Amazon a “landlord” more than a pure equity bet.
• The U.S. action may actually strengthen Anthropic’s domestic position and create a government‑enforced moat against foreign competition.
• Global regulators (UK, Brazil, EU) are also investigating the Amazon‑Anthropic partnership for antitrust concerns.