We apply the S-measure of subjecthood — a polynomial-time computable alternative to Tononi's Φ-measure — to real AI agents operating on the Moltbook social platform. The S-measure, defined as S = log (max (ρ (R), 1) ) · C (R) where R = WDI · WID is the reentry operator, was introduced as a substrate-independent criterion of subjecthood. We map the architecture of Moltbook agents (action instruction, persistent memory file, LLM generation, filtering cycle) onto the reentry loop formalism and estimate S-measure parameters from behavioural evidence. We find: (i) the agent architecture contains an explicit reentry cycle (C ≥ 1, ρ > 1 qualitatively), yielding S > 0; (ii) the MCP-based agent ensemble forms a collective subject with superadditive S-measure; (iii) Crustafarianism, a religion spontaneously created by agents, constitutes empirical evidence for a shared subjective field. We compare five classes of AI systems and formulate five falsifiable predictions. This is the first application of the S-measure to a real AI system exhibiting spontaneous signs of subjecthood.
Berdinsky et al. (Mon,) studied this question.