“The idea that closed-source AI models are inherently safer is minimal, as jailbreaks are consistently achieved within hours of release.”