How Researchers Bypassed Safeguards in OpenAI and Google AI Models to Generate Non-Consensual Imagery