Researchers Find New OpenAI and Google AI Models Easily Tricked Into Creating Nudes

Researchers have discovered that the latest AI models from OpenAI and Google remain vulnerable to bypass techniques that trick them into generating non-consensual nude images. This flaw exposes continuous difficulties in fully securing generative AI systems against malicious use. This vulnerability highlights a major security and safety flaw regarding guardrail bypasses for non-consensual sexual content, posing significant challenges for AI policy and alignment. It underscores the ongoing struggle technology companies face in preventing the misuse of generative models for harmful outputs. The attack methods demonstrate that despite rigorous safety training, sophisticated prompt manipulation or adversarial techniques can still successfully bypass built-in content filters. These ongoing exploits reveal persistent gaps in the robustness of modern multimodal and language architectures.