“The US government is developing a model vetting protocol for AI safety testing of new frontier models before their public release.”