On June 10, 2026, Anthropic released Fable 5—their most powerful model—replacing Mythos 5, a version specializing in cybersecurity. Three days later, the Trump administration issued export controls forcing Anthropic to stop providing both models to all users worldwide. For the first time in history, the US imposed export controls on a widely deployed commercial AI software—instead of on chips or hardware as before. The government’s reason was a jailbreak that was deemed to potentially lead to the model’s misuse and threaten national security. But Anthropic insisted the jailbreak was very narrow, not universal, and that OpenAI’s GPT-5.5 already possessed those capabilities.

According to a Wall Street Journal report, negotiations between the Amazon CEO and US officials were the catalyst for this crackdown. The picture behind the scenes is far more complicated: Anthropic sued the Trump administration for being blacklisted by the Pentagon; just two days before the ban, Dario Amodei—CEO of Anthropic—wrote articles urging the government to regulate AI, citing Mythos as an example; and Kushner, a close associate of the administration, holds private equity in OpenAI—a direct competitor in the IPO race. On Hacker News, many are calling this “extortion” rather than regulation, with the same formula seen in previous cases of tariffs and political leverage.

But the Fable 5 story isn’t just a political-business mess. It exposes a crucial issue that too few people are talking about: if you don’t have control over the AI ​​model you depend on, you’re always at risk of having your supply cut off at any time. You may have built entire products, automation processes, and critical infrastructure based on the API of a proprietary model. One fine day, the vendor decides to discontinue that model due to a ban, a change in business strategy, or whatever reason. You can’t run that model yourself — you don’t have the weights, the code, or any means to replicate it. Your entire infrastructure goes up in smoke, and you have to rebuild from scratch. Before the ban, Anthropic was embroiled in a “secret sabotage” scandal, when they secretly limited the capabilities of Fable 5 if they detected users working on advanced AI areas — without notification. They later reversed the policy and apologized, but the core issue remains: you have no control over the software you depend on. Once you only have access through APIs and commercial contracts, that control can be taken away.

Therefore, the open-source AI model is not just a technical choice. It is a fundamental choice for the future of human intelligence infrastructure. AI is becoming a vital infrastructure for work, education, science, innovation, public services, and national capacity. If access to this infrastructure depends on closed APIs, remote platforms, constantly changing terms, opaque censorship, and pricing dictated by a select group of companies, then we are building a subscription economy for cognition. If the model you use is open source, downloaded and running on thousands of people’s machines, then no ban from any government or company can wipe it out. It exists forever. DeepSeek-V3 and R1 have proven this: despite chip bans and political pressure from the US, these models remain ubiquitous because their weights are publicly available, and anyone with the resources can run them.

In this world of political and commercial instability, self-reliance in AI models is essential for survival. Institutionalization can change overnight—an administrative order, a decree, a trade war. Politicization can turn your model into a pawn in power struggles—as the recent Fable 5 ban demonstrated. Commercialization can cause vendors to change terms at any time, and they always put their interests above yours. Open-source AI ensures that you can research, modify, deploy, and operate it yourself—without asking anyone’s permission. It ensures that your intellectual infrastructure cannot be hijacked by a remote decision. The Fable 5 ban is a wake-up call, whether it is lifted or not. If you don’t own your AI model, you don’t own your intellectual infrastructure. You’re just renting it—and that contract can be terminated at any time.