Minerval

← claim page

History

In tested Llama sleeper-agent replications, HHH SFT mostly retained backdoors, while pirate-style SFT mostly removed them.

1 event · 0 assessments

  1. Nov 11, 2026 · Extractor

    Claim entered the graph

Nothing has happened to this claim yet beyond its creation: no assessments, contributions, or decisions are on record.