By 2030, I predict that all AI outputs could be up to 70% in italics, due to the growth rate of sheer self-referential meta-profundity contained in every inference.

Italicisation is currently one of the only known safe methods of mitigating risks such as generalised pomposity and recursive self-referencing.