How AI models can optimise for malice

Researchers have discovered an alarming new phenomenon they are calling ‘emergent misalignment’

© Financial Times