Researchers find new AI models from OpenAI and Google remain vulnerable to generating non-consensual explicit content

A recent study co-authored by Google researchers highlights the growing challenge of preventing the creation of non-consensual explicit content using generative AI tools. Despite increased scrutiny, researchers warn that as AI image generators from industry leaders like OpenAI and Google become more advanced, they are paradoxically becoming more susceptible to being tricked into producing harmful, non-consensual sexual imagery. The study, which surveyed users across the US, Australia, and the UK, emphasizes that while explicit content is often distributed through private channels, the most effective intervention point remains the source models themselves. Red-teaming efforts conducted in recent months indicate that current safety guardrails are struggling to keep pace with the evolving capabilities of these models. This ongoing issue underscores the tension between rapid AI development and the implementation of robust safety protocols, as regulators continue to face challenges in effectively curbing the spread of AI-generated abuse.