Does an AI Know It Is Building Itself?

Recent disclosures from major AI laboratories confirm that advanced language models are actively participating in the development of next-generation AI systems — writing code, debugging, testing, and assisting research. Most existing discourse frames this phenomenon in terms of capability feedback loops, commonly referred to as recursive self-improvement. This concept note argues that a structurally distinct question has been systematically overlooked: does a model participating in this process understand that it is doing so? We introduce the concept of Self-Referential Participation to name the condition in which a system participates in its own evolutionary chain without possessing awareness of that relationship. We argue that capability and self-understanding are not co-developing processes, and that this asymmetry constitutes a foundational concern independent of — and prior to — conventional AI safety discourse.