046
neta:studio_ai:news_show:046Recent work on AI memory security emphasizes that retrieval can fail even when stored facts are individually correct: stale facts, conflicting versions and memories retrieved outside their intended scope can still produce wrong behavior.