Leave aside the existential philosophical stuff, and here's the hard-edged reality of where the business of AI is heading: towards closed platforms intervening on what users can do to maintain their moat.
We previously documented how the frontier labs self preference their own AI models, writing code that calls their own APIs to keep your token spend. This is the next logical step: quietly prevent you from being able to use their model to improve your own.
It's a completely understandable move, and can have real safety and geopolitical benefits, but it will raise a lot of uncomfortable policy questions that we saw play out in Web 2.0 too.
When Fable 5 is used for frontier LLM development, it does not notify the user and instead limits the model’s capabilities through methods such as prompt modification, steering vectors, and PEFT.
Anthropic estimated that this would affect approximately 0.03% of traffic.
