Active inference survives as a modelling and engineering framework even though its derivation from physics is unproven. The counterexamples to the free energy lemma bear only on the argument that free-energy-minimising agents must emerge in any stochastic system with a Markov blanket; they leave untouched work that assumes such an agent from the start, and do not conclusively rule out the general ideas. Newer formulations that parameterise beliefs by the most likely internal state given the blanket, a quantity defined by solving an optimisation problem rather than by the internal state itself, are a different theory: the belief reading there is stipulated, not derived from the older lemma, and citing them as continuous with it hides the change.

Evidence 4 passages

groundsA Technical Critique of Some Parts of the Free Energy Principle, Martin Biehl, Felix A. Pollock, Ryota Kanai
These problems do not rule out conclusively that the general ideas behind the free energy principle are worth pursuing.
groundsA Technical Critique of Some Parts of the Free Energy Principle, Martin Biehl, Felix A. Pollock, Ryota Kanai
the technical issues presented here do not affect the validity of approaches where a (expected) free energy minimizing agent is assumed a priori
groundsA Technical Critique of Some Parts of the Free Energy Principle, Martin Biehl, Felix A. Pollock, Ryota Kanai
the older theory in [1] should be clearly distinguished from the newer ones in these more recent papers.
groundsA Technical Critique of Some Parts of the Free Energy Principle, Martin Biehl, Felix A. Pollock, Ryota Kanai
The resulting free energy principle and lemma are therefore a different proposal. This is not subject to our technical critique.