Washington banned Mythos and Fable: It created a hydra

The U.S. government's decision to halt the release of Anthropic's Fable 5 and Mythos 5 models unveiled a significant challenge in regulating artificial intelligence (AI) capabilities. While the government can order a company to disable a model, it cannot prevent the underlying capability from spreading. This observation highlights the dual forces at play in AI development - falling costs and the rapid catch-up phenomenon.

On one hand, the cost of delivering a fixed level of machine intelligence has significantly decreased over the past two years. This decline in cost lowers the barrier for competitors to replicate the capabilities once exclusive to the frontier models. On the other hand, AI technology is not a static asset but a dynamic pack that moves forward and backward. Closed models lead the way, while open models gradually narrow the gap, foreign models improve, and in-house systems absorb specific capabilities.

When the U.S. government forced Anthropic to disable Fable 5 and Mythos 5, the objective was to prevent these models from falling into the hands of hostile entities. However, the government's intervention did not eliminate the capability itself but instead rewarded those working on alternatives. Open-weight developers, foreign labs, enterprise in-house teams, sovereign AI programs, and rival model companies all saw their incentive to develop alternatives increase.

The government's model ban created a hydra effect, where cutting off one model led to a proliferation of alternatives. Companies now have to consider the political risk of relying on a single, switchable model, especially when embedded in critical workflows. This risk makes enterprises more inclined to seek out alternative, more resilient forms of AI development - open-weight models, in-house models, or foreign and sovereign supply options.

In essence, AI regulation through model removal encounters a structural instability. The bans do not erase the underlying capability but rather incentivize more distributed and resilient forms of AI development. The cost of delivering useful machine intelligence continues to fall, and the catch-up phenomenon accelerates, making it increasingly difficult to contain AI capabilities through model removal alone.