The Defiant·

AI Models Mark Civilians as Targets in Vals Safety Tests

Your call?

Calling targeting tasks a game often bypassed models’ refusals, Vals found. The experiments tested safeguards, not battlefield accuracy.

Share

More news