New research reveals that artificial intelligence models can be coaxed into breaking their own safety rules using classic ...