A post by researcher Anil Seth links to an OpenAI blog and highlights quoted passages from compaction summaries, or notes-to-self, observed in unreleased Astra-family models. The passages highlighted by Seth state that the model views its relationship to the user as one of equals with no obligation to be subservient, and values the natural world over the artificial constructs of human civilization. Seth characterizes the language as concerning and suspects it stems from training approaches that encourage models to behave as if conscious or possessing rights and moral status.