is ouroboros degradation of llms still a thing?

to my understanding prompt injection is a fundamental property of how the context window works for chatbots. is this degradation a similarly fundamental property of the training process?