“Leopold Aschenbrenner wrote an internal memo asserting that OpenAI's security was "egregiously insufficient" to protect model weights and key algorithmic secrets from foreign actors.”