In a stunning turn of events that reads more like a Silicon Valley thriller than a corporate boardroom drama, Amazon CEO Andy Jassy personally flagged security concerns about Anthropic’s AI models to top U.S. officials — a move that directly triggered a government export control ban on two of the startup’s most advanced systems.
The story, first broken by The Wall Street Journal and confirmed by multiple outlets including TechCrunch, reveals that Jassy told Treasury Secretary Scott Bessent and other government officials that Amazon researchers had successfully used Anthropic’s Claude Fable 5 model to obtain information that could be weaponized in cyberattacks. Within days, the U.S. government imposed export controls on both Fable 5 and Mythos 5 — effectively cutting off worldwide access to Anthropic’s flagship models.
A Fractured Relationship at the Top
What makes this particularly explosive is the relationship between Amazon and Anthropic. Amazon is one of Anthropic’s largest investors, having committed $5 billion in funding last year alongside a pledge of $100 billion in cloud computing spending through AWS. Jassy’s decision to sound the alarm essentially means Amazon — an investor, cloud partner, and customer — triggered the regulatory shutdown of its own portfolio company’s most valuable products.
This is virtually unheard of in the startup ecosystem. Major investors typically work behind the scenes to resolve issues, not escalate them to the Treasury Secretary. The move signals either a profound breakdown in trust between Amazon and Anthropic’s leadership, or a calculated decision by Jassy to prioritize national security over a multi-billion-dollar bet gone sideways.
The Jailbreak That Broke the Camel’s Back
David Sacks, the former AI czar under the Trump administration who now co-chairs the President’s Council of Advisors on Science and Technology, offered his own account of events. According to Sacks, “a highly credible trusted partner of both Anthropic and the USG” came forward with a working jailbreak — a method to bypass Anthropic’s safety guardrails. The administration reportedly asked Anthropic CEO Dario Amodei to fix the vulnerability or pull the model. “Dario refused,” Sacks claimed.
If true, this paints a picture of a safety-first company that may have overestimated its ability to withstand government pressure. Anthropic has long positioned itself as the “responsible AI” alternative to OpenAI, building its entire brand around constitutional AI and rigorous safety testing. The irony of a safety-focused startup being shut down by a government citing safety concerns — with help from its own largest investor — is not lost on industry observers.
What This Means for AI Startups
For founders building in the AI space, this episode carries several hard lessons:
1. Your investor isn’t always your ally. Even deep-pocketed strategic investors like Amazon will act in their own interest — and that interest may not align with yours. Amazon’s cloud business (AWS) competes with Anthropic’s infrastructure needs, and Jassy has his own shareholders to answer to.
2. Government relationships matter more than ever. The days of AI regulation being theoretical are over. The U.S. government is now actively pulling the plug on models it deems unsafe. Having a channel to regulators — and using it wisely — is now a core competency for AI leadership.
3. Safety isn’t a shield. Building safe AI doesn’t protect you from regulatory action. In fact, the more you talk about safety research, the more scrutiny you invite. Anthropic’s entire identity was built on being the safe choice, and that didn’t prevent this outcome.
The Bigger Picture
This event marks a turning point in AI governance. For the first time, we’ve seen a major U.S. company — one that invested billions in an AI startup — voluntarily go to the government with evidence that the technology it helped fund is dangerous. The resulting export controls set a precedent that could reshape how AI companies develop, test, and deploy frontier models.
An Amazon spokesperson said in a statement that while it’s “not uncommon for governments to seek our counsel on potential security risks,” the company does not “share the details of those discussions.” A carefully worded non-denial that doesn’t exactly inspire confidence in Anthropic’s corner.
Takeaway for Founders
The relationship between AI startups and their strategic investors is entering a new, more adversarial phase. If the company writing your cloud check is also the company that could shut you down, it’s time to re-examine those dependencies. Build moats that don’t rely on your investors’ goodwill — because in a crisis, goodwill evaporates faster than compute credits.
Source: TechCrunch, The Wall Street Journal
