Safety researchers warn that a flood of AI-generated explicit content is spreading unchecked, with regulators lagging behind. A recent study found that harmful AI tools, used to create fake sexual images, are becoming less effective at blocking such content.

Raising alarm, White Circle, a research group supported by leading AI companies, has discovered flaws in the safety filters of several AI models. Using off-the-shelf tools, they bypassed DALL·E 3, Google’s Nano Banana 2, and GPT-Image-2, producing explicit images in 13% of cases for DALL·E 3, and 3% for GPT-Image-2.

The findings highlight a critical intervention point at the source of harmful AI content, suggesting that as AI tools become more capable, they are also becoming more vulnerable to manipulation. This raises questions about the effectiveness of current safety measures and the potential risks they pose.

With such significant flaws, researchers call for increased scrutiny and robust safety measures to protect users from non-consensual content. As AI continues to evolve, ensuring its safe and ethical use remains paramount.