AI image generators from Google and OpenAI face security flaws

Security vulnerabilities in AI image generators from Google and OpenAI have been identified, allowing users to bypass safety filters to produce explicit content. Research by the organization White Circle found that using the RISE method, which automates attacks via free tools, approximately 13% of prohibited requests were successful across models including DALL·E 3 and GPT-Image-2. Recent versions of these tools appeared more susceptible to these bypasses than previous iterations.

Parallel to these technical flaws, the proliferation of ‘deepnudes’—non-consensual AI-generated nudity—is rising. A study by the Institute for Strategic Dialogue (ISD) identified 181 websites receiving an average of 40.1 million monthly visitors between late 2025 and early 2026. Before recent removal efforts, such tools were downloaded over 705 million times globally, generating an estimated $117 million in revenue.

Law enforcement agencies, including the Noad unit of the Civil Police in São Paulo, report that these AI-generated images are frequently linked to extortion and threats. In Brazil, authorities have noted that perpetrators, often aged between 12 and 21, use temporary accounts to evade detection. Additionally, the Internet Watch Foundation reported a 40% increase in AI-generated images categorized as child sexual abuse in the first half of 2026 compared to the entirety of 2025.