Altman confirms OpenAI is running systems that can self-improve, and has created a Head of Preparedness to manage it. The corpus has followed this from Codex watching its own training on the 15th, through a leaderboard for models post-training models on the 18th, to an org chart position on the 28th.
The conversions are the other story. Eric Raymond and Eliezer Yudkowsky — from opposite ends — both conclude it has happened, the latter after Opus 4.5 evaluated its own personhood against historical definitions.