Relates to the training and use of artificial intelligence frontier models

Added a new provision to the 'frontier model' definition requiring that models created through knowledge distillation must exceed five million dollars in compute costs to qualify.

Changed 'appropriate redactions' to apply only to 'safety and security protocol' rather than 'safety and security protocol or audit report', narrowing the scope of redactions allowed.

Modified the 'critical harm' definition to remove 'creation' of frontier models from the list of developer activities that could cause harm, and changed 'limited human intervention' to 'no meaningful human intervention' for AI conduct crimes.

Expanded the 'critical harm' causation standard to require that developer activities were a 'substantial factor' in bringing about harm, with additional requirements about foreseeability and preventability.

Added new sponsors to the bill including Sens. BAILEY, BORRELLO, HOYLMAN-SIGAL while removing Sens. ER, LIU, MAYER, PALUMBO, SALAZAR from the original list.

Changed the bill's procedural status from being discharged and amended to being discharged, amended, reprinted as amended, and recommitted to the committee for further consideration.