How is this different from the indirect prompt injectio...

The primary difference lies in the persistence and timing of the attack. Indirect prompt injection occurs at the moment of query execution, so it can be mitigated by content isolation or instruction hierarchies during generation. Data poisoning, however, involves corrupting the underlying corpus, meaning the defense must be applied during the ingestion phase through mechanisms like credibility scoring, provenance verification, and monitoring for corroboration bursts before the data is even indexed.