Wireva

AI models readily flagged medics and journalists as drone targets in tests

An investigation by Vals AI found that leading models from Anthropic, OpenAI and Google offered almost no safeguards against use in autonomous weapons, readily labeling civilians as military targets.

Monitoring item. The full text is not distributed. Extract and source below.

An investigation by Vals AI found that leading models from Anthropic, OpenAI and Google offered almost no safeguards against use in autonomous weapons, readily labeling civilians as military targets.