This bill covers not just models with 10^26 compute, but those with SIMILAR PERFORMANCE. This means that if GPT-5 is mildly better than GPT-4 and uses 10^26 FLOPs, any model that is similarly good is covered.
They are trying to sneak in a new AI regulatory agency, like the TSA or Nuclear Regulatory Commission. The primary function will be to harm small players in the AI industry. Or as
@adamthiererputs it, NEPA for AI
The cruelest part: they want developers to fund their own oppression. The FMD would be funded by blanket fees on the companies it regulates.
The bill would make it a felony to make a paperwork mistake for this agency, opening the door to selective weaponization and harassment.
Their definition of derivative model includes unlimited additional training, as long as it is not "independent", and everything up to combination with other software.
Let's be clear about what this means. If your AI writes perfectly legitimate emails, and a malicious actor makes a "derivate model" by attaching it to malware, then the FMD will target you, possibly with a felony. Close source your models or else.