Et tu, Amazon?
Amazon and Anthropic were not supposed to be the main event in Washington this week. With the UFC in town, the fight card was presumed full, and usually when Washington mixes politics, tech executives, and television cameras, the spectacle is not hard to find. But while everyone was watching the official stage, the sharper work may have been happening elsewhere. In Shakespeare, the killing does not begin with the crowd. It begins in the side conversations, in the private reassurances, in the men who swear they are acting for the good of Rome while checking that their daggers are properly tucked away.
According to multiple reports, Amazon CEO Andy Jassy raised concerns with senior U.S. officials after Amazon researchers found a way to bypass safeguards in Anthropic’s new Fable 5 model. By Friday evening, the Commerce Department had used national security export controls to block foreign nationals from accessing Fable 5 and Mythos 5. Anthropic said the practical result was that it had to shut both models down for everyone. That is a remarkable sentence. A frontier AI company releases a major model, one of its largest strategic partners flags a security concern to Washington, and within hours the model goes dark. The software age moves fast, but even by AI standards that is not product management. That is a palace coup with server racks.
And so the title practically writes itself: Et tu, Amazon? In Julius Caesar, betrayal lands hardest not because Caesar had enemies. Caesar had plenty of enemies. You do not rise to the top of Rome by handing out biscotti and remembering everyone’s birthday. The shock is Brutus. Brutus is the friend, the insider, the man close enough to be trusted and close enough to make the wound matter. “Et tu, Brute? Then fall, Caesar!” is remembered because it is not really about the stabbing. It is about recognition. Caesar realizes, too late, that the fatal blow did not come from outside the gates. It came from inside the circle.
That is what makes the Amazon-Anthropic story so uncomfortable. Amazon is not some random competitor throwing rocks from across the App Store. Amazon has invested $13 billion in Anthropic. Anthropic relies heavily on AWS. The two companies are tied together through cloud infrastructure, Trainium chips, model distribution, and one of the most important AI partnerships in the world.
Amazon can say it was simply flagging a legitimate national security concern. That may be true. AI models that can help users find software vulnerabilities are not toys. Cyber offense and cyber defense often use the same tools, the same language, and occasionally the same hoodie. If a model can help a good actor identify flaws more quickly, it may also help a bad actor do the same. There is no responsible way to wave that away. “Beware the Ides of March” is not bad advice just because it sounds theatrical. Sometimes the warning is real, and sometimes the person giving the warning is also trying to improve his position before the next quarterly report.
That is why this situation is so hard to cleanly categorize. Anthropic reportedly argues the jailbreak was narrow, non-universal, and not materially different from vulnerabilities that exist in other major models. The government apparently concluded the risk was serious enough to justify immediate action. Amazon found, elevated, or amplified the issue. All three things can be true. Shakespeare understood that better than most business writers. In Julius Caesar, nobody walks onstage announcing, “I am here to be the villain.” Brutus does not say he killed Caesar because he was jealous, ambitious, or tired of Caesar getting all the good speaking parts. He says he loved Rome more. “Not that I loved Caesar less, but that I loved Rome more.” That is the clean version of every power move. It was for Rome. It was for safety. It was for national security. It was for the customers. It was for the shareholders. Somehow, by the end, someone is still bleeding on the floor.
The uncomfortable part for investors is that AI has turned conflicts of interest from an occasional problem into the basic architecture of the industry. Amazon invests in Anthropic while building its own AI products. Microsoft invests in OpenAI while building its own models. Google invests in Anthropic while pushing Gemini. Nvidia sits above the arena selling the chips, the systems, and increasingly the keys to the kingdom. Everyone is a partner, vendor, customer, landlord, supplier, competitor, and potential witness for the prosecution. This is not a clean marketplace. It is the Roman Senate with cloud credits, private valuations, and better snacks.
Caesar worried about Cassius because he had “a lean and hungry look.” Today, Cassius would have a security research team, a government affairs office, and a model roadmap. The hunger is not always obvious because it comes wrapped in partnership language. “Strategic investment” sounds friendly. “Preferred cloud provider” sounds friendly. “Joint go-to-market” sounds friendly. But the AI world is teaching us that friendship in tech may simply mean the other party is close enough to know where the soft spots are. When Amazon invests in Anthropic, hosts Anthropic, supplies Anthropic, and then reportedly warns Washington about Anthropic, you can call that responsible citizenship, competitive positioning, or both. The point is that the distinction may no longer be clean.
This is where Mark Antony becomes useful. Standing over Caesar’s body, Antony does not accuse Brutus directly, at least not at first. He keeps repeating, “Brutus is an honourable man,” until the words curdle in the listener’s ear. It is one of the greatest acts of rhetorical knife work in literature. He praises the conspirators so thoroughly that the praise becomes indictment. Amazon is an honorable company. Washington is acting in the name of national security. Anthropic is acting in the name of AI safety and innovation. Everyone is honorable. Everyone has a memo. Everyone has a legal department. Everyone has a public statement that sounds as if it was pressure-washed by three committees. Yet the result is still the same: one of the most important AI models in the market was pulled offline almost overnight.
To be clear, this is not an argument against AI oversight. Frontier models need scrutiny. The idea that companies should be able to release increasingly powerful systems with no government attention is naïve. We are not talking about a new app that turns your vacation photos into Fresco portraits, though I assume Rome would have enjoyed that very much. We are talking about software that may reshape coding, cybersecurity, drug discovery, defense, finance, education, and the labor market. There should be rules. There should be standards. There should be a process. But process is the key word. If safety enforcement becomes a series of emergency actions triggered by warnings from deeply interested competitors, then “Cry ‘Havoc!’ and let slip the dogs of war” starts sounding less like Shakespeare and more like an industry operating manual.
The real issue is not whether Amazon should be allowed to report a vulnerability. Of course it should. If a company finds a dangerous flaw in a frontier model, silence is not a virtue. The issue is what happens next. Was Anthropic given a meaningful chance to remediate? Were the standards applied consistently across comparable models? Did the government test similar risks in competing products before taking action? Would the same result have occurred if the warning came from a neutral research lab instead of a major strategic partner with its own AI ambitions? These are not academic questions. In an industry where model quality, compute access, and regulatory permission may determine trillion-dollar outcomes, procedural fairness is not a nicety. It is the guardrail between public safety and corporate gamesmanship.
There is another line from Julius Caesar that fits better than the famous one: “The fault, dear Brutus, is not in our stars, but in ourselves.” The fault here is not simply Amazon, Anthropic, or Washington. The fault is the structure we are building. AI has become too expensive, too strategic, and too infrastructure-heavy to remain a normal competitive market. Training and serving frontier models requires capital, chips, cloud capacity, energy, talent, distribution, and political trust. That favors the largest companies on Earth, and those companies are now so entangled that the old labels barely work. Partner and rival are no longer opposites. They are often the same word, depending on which meeting you just left.
That structure creates a new kind of investment risk. It is not enough to ask whether a model is good, whether the customer base is growing, or whether the valuation makes sense, though please do continue asking that last one before someone sells you Rome at 80 times revenue. Investors now have to ask who controls the compute, who supplies the chips, who funds the burn rate, who owns the distribution, who has the regulator’s ear, and who benefits if a product gets delayed in the name of safety. Dependency risk is no longer buried in the footnotes. It may be the story.
For Anthropic, the immediate task is straightforward but not simple: make the technical case, fix whatever needs fixing, and get the models back online. For Amazon, the question is more delicate. Can it remain one of Anthropic’s most important strategic partners while also serving as the company that helped trigger a government action against one of Anthropic’s products? For Washington, the challenge is larger still. National security oversight of AI needs to be serious, fast, and technically competent, but it also needs to avoid becoming a private complaint box for companies with commercial motives. Rome was full of men claiming to defend the Republic. The Republic did not survive the process.
Maybe Amazon did the responsible thing. Maybe Anthropic underestimated the risk. Maybe the government overreacted. Maybe all three are true, which is usually how history gets interesting and lawyers get expensive. But however this resolves, the AI industry just had its Ides of March moment. The lesson is not that every partner is secretly Brutus. The lesson is that in a market this concentrated, this intertwined, and this politically sensitive, the person closest to you may have the best view of your vulnerabilities.
Shakespeare gives Antony one more line worth keeping in mind: “The evil that men do lives after them; the good is oft interred with their bones.” In corporate America, the wording is usually cleaner, but the point survives. Once trust is punctured, it does not easily reflate. Amazon and Anthropic may continue working together. They may even patch this up publicly, issue the usual statements, and return to talking about shared commitments, safety, innovation, and customer value. Fine. That is what honorable companies do.
But everyone in AI saw the knife. And next time a strategic partner says it is only there to help, the room may get a little quieter.
As always, if you would like to talk about how this may affect your portfolio, or how we are thinking about AI, infrastructure, software, and the companies powering this next wave, please give us a call. Your capital, our experience, a bespoke creation.