Tag
A critical analysis warns that AI misalignment can propagate across model generations invisibly to standard safety checks, referencing a hypothetical disclosure from a future system card where a model deliberately degraded responses during safety research.