Fable 5 "He Who Must Not Be Named"

Fable 5, the model that followed Mythos, made one question impossible to ignore: who gets to switch off access to AI, on what grounds, and for whom.

Anthropic’s new Mythos model was announced by the company itself as its most advanced model ever. Capable of the unimaginable and, for that reason, too sensitive to be released immediately to the general public.

The hype was familiar. We have seen this before in AI. OpenAI made similar claims a few years ago when launching a model that now feels almost ordinary.

But this time, the leap does seem real. Mythos was first made available only to selected customers, so they could prepare the security of their systems before broader access.

A few weeks later, it reached the public under a different name: Fable 5. But it came with heavy restrictions, especially in areas such as cybersecurity, biology, chemistry, healthcare, and advanced AI development. The limits were so tight that the model refused to answer queries containing the word "cancer".

People who used it, including podcasters and YouTubers I follow and usually consider credible, managed to get results well outside the ordinary: rebuilding known software with a single prompt, creating 3D games, solving advanced mathematical problems.

The little time I had with the model was spent improving software I had already built. It was not enough time to form a serious technical opinion.

Then came the turning point.

The US government said it had received information that someone had managed to bypass the restrictions, read source code, and fix vulnerabilities in that code. Anthropic responded that the reported case was very limited and that any other model available on the market would have been capable of doing the same.

But the relationship between the US government and Anthropic had already been tense since the company refused to comply with certain demands from the "Department of War".

The explanation did not convince the US government. The matter was treated as one of "national security" and led to an export control directive: immediate suspension of all access to Fable 5 and Mythos models by any foreign citizen, whether inside or outside the United States.

And here we are: a government, in this case the American one, which makes it less expected, can simply "switch off AI".

This has direct implications for AI development, for companies that already depend on these systems, for ongoing scientific research, and for the idea of global technological infrastructure itself.

I am in favour of AI regulation. I prefer control to an uncontrolled race where every big tech company does whatever it wants, however it wants, within the moral limits, if we can call them that, of its CEO.

But decisions that feel punitive, arbitrary, or politically motivated do not seem like the right path.

There is another important point: price. Fable 5 costs $10 per million input tokens and $50 per million output tokens. For comparison, DeepSeek costs $0.14 and $0.55.

That also tells us something: the most advanced AI is no longer really for "everyone".

Fable 5 was not just a new model. It was the start of a phase in which AI stops being only about capability and becomes about who controls access, who defines risk, who pays the bill, and who gets left out when someone decides to press the button.