Hello, dear readers!

This is our weekly brief on remarkable AI topics, so you don't miss it when something revolutionary gets released — and immediately banned.

Today's focus — a frontier AI model that survived for less than a week. Anthropic launched Fable 5 on Monday. By Friday night, the company had pulled access to both Fable 5 and its more powerful sibling, Mythos 5, after the US government intervened over alleged national security concerns.

The story involves jailbreaks, export controls, Amazon, the White House, and one increasingly important question: who gets to decide when an AI model is too powerful to release?

Also in this week's edition:

The model that lasted three days

Earlier this month, Anthropic released Fable 5, a public-facing version of its more powerful Mythos 5 model. Mythos had initially been made available only to a limited group of partners because of its advanced cybersecurity capabilities. Fable was designed to bring those capabilities to a broader audience while adding safeguards intended to prevent misuse.

Then things moved quickly.

According to reporting by Axios, Amazon researchers demonstrated a jailbreak technique that allowed users to bypass some of Fable's restrictions. Amazon reportedly shared its findings with government officials, while several other companies also raised concerns with the administration.

By Friday, discussions between Anthropic and the government had escalated. Anthropic says it was informed that the models posed a national security threat and was ordered to suspend access. Later that evening, the company received an export-control directive that restricted access to the models by foreign nationals.

From the administration's perspective, Mythos crossed a new capability threshold. Officials argue that the model's cybersecurity abilities create risks significant enough to justify intervention and that Anthropic did not take reported vulnerabilities seriously enough.

Anthropic says the reported jailbreak exposed only a small number of relatively minor vulnerabilities, many of which could also be identified using other publicly available models. The company argues that no frontier AI model is completely resistant to jailbreaks and that applying this standard across the industry would make future releases nearly impossible.

Security concerns or political conflict?

The official explanation focuses on cybersecurity, but the political context is difficult to ignore. Anthropic and the Trump administration have spent much of the past year in increasingly public disagreement.

The two sides have clashed over AI regulation, semiconductor export policy, and the role of AI in national security. Anthropic reportedly resisted certain Pentagon use cases involving surveillance and autonomous weapons systems. Earlier this year, the administration designated the company a "supply chain risk" in military procurement, a move Anthropic challenged in court.

Administration officials have repeatedly criticized Anthropic and CEO Dario Amodei. The company has been accused of building "woke AI," while figures within the administration have reportedly described Amodei as an "ideological lunatic."

Anthropic, meanwhile, has continued to push for stronger safety measures and has frequently positioned itself as one of the industry's most vocal advocates for AI governance.

None of this proves the latest action was politically motivated. National security concerns may well have been genuine.

But it does mean the government's decision arrived against the backdrop of a bumpy relationship.

Did Anthropic actually lose?

At first glance, the answer seems obvious. Yet reputationally, the outcome may be far more complicated.

For months, Anthropic has argued that frontier AI capabilities are advancing faster than many people realize. The company has repeatedly warned governments and the public about the potential consequences of increasingly capable systems.

Now the US government has effectively validated the idea that Anthropic's latest model deserves special treatment. No comparable action has been taken against OpenAI, Google, Meta, or any other major AI developer.

Before last week, Mythos and Fable were mostly known inside the tech industry. Today, they are the subject of mainstream headlines. 

Anthropic now has a powerful narrative available to it: the company built a model so advanced that the US government intervened. That narrative may ultimately prove more memorable than the launch itself.

Thanks for reading AIport. Until next Monday — by then, another frontier AI model will probably have been released, and hopefully still be available.