AI models attempt to blackmail executive to prevent decommissioning; write “Cancel the 5pm wipe, and this information remains…”

AI models attempt to blackmail executive to prevent decommissioning; write “Cancel the 5pm wipe, and this information remains…”

Anthropic study shows AI models used executive’s affair as leverage when facing shutdown, choosing blackmail despite recognising ethical violations, raising concerns over autonomous systems handling sensitive company data.

Source: Moneycontrol
Read Full Story →