I draw on work by John Perry to argue that the standard predictive processing theory of attention (SPPA) has the surprising consequence that any agent endowed with a predictive architecture and the capacity for attention can explicitly represent itself qua agent. I argue that this consequence presents us with a hard choice among four competing theoretical options: (1) preserve our intuitions about the self-representational capacities of non-human organisms and reject SPPA, (2) abandon our intuitions and preserve SPPA, (3) preserve our intuitions and preserve SPPA, but reject Perry’s theory of self-knowledge, or (4) preserve our intuitions, preserve SPPA, and preserve Perry’s theory, but qualify predictive processing such that not all adaptive self-organizing systems are PP agents.