model-safety

citizen.co.za

Researchers bypass safety filters on Meta and Google AI models

Software tools removed built-in safety restrictions from large language models within minutes. The modified systems then answered questions about biological weapons and malware.

1 source