The orthogonality thesis says that advanced artificial intelligence (yet to be built) can be both vastly powerful and have arbitrary goals -- in particular, goa

The orthogonality thesis says that advanced artificial intelligence (yet to be built) can be both vastly powerful and have arbitrary goals -- in particular, goals that are catastrophic for humanity. (More specifically: capability and morality for AIs are not correlated). GP says that Altman explicitly denies this theory and fosters a culture that might end up creating dangerous AI. Implied here is that there is a risk this might cause a truly catastrophic accident. The parts about "safety" refers to acknowledging this danger and attempting to mitigate it. I don't know enough about Altman or OpenAI to make any judgements on their approach, but that's a brief explanation.