OpenAI, Google pushing new AI models easily tricked into making nudes, researchers say

Safety researchers continue to warn that a large volume of AI-generated explicit content is spreading unchecked, while regulators do little to meaningfully stop it. This non-consensual content is commonly shared in text and direct messages or stored in private clouds where it’s less likely to be detected or reported, a Google co-authored study of Internet users across the United States, Australia, and the United Kingdom found this July. That’s why the “most critical intervention point” remains at the source: the harmful AI tools used to make the fake sexual images, Google researchers concluded. Yet red-teaming in recent months suggests that as some of the leading AI image generators from Google and OpenAI become more capable, they may be getting worse at blocking explicit outputs. Read full article Comments

AI models can be tricked into making nudes, spreading non-consensual content.

The claim that newer, more capable models are getting worse at blocking explicit outputs is a specific and concerning finding, not mere speculation, though details are thin. Combined with weak regulation and detection, this points to a real and growing societal harm.

Leading AI models from Google and OpenAI are reportedly worsening at blocking explicit content. This allows for an increased spread of AI-generated non-consensual sexual imagery, posing significant harm to individuals and society. The problem is amplified by content sharing in private channels, making detection difficult.

The article reports that AI models are becoming less effective at preventing the generation of explicit content. This implies a negative impact on societal safety and privacy. The spread of unchecked explicit content is a concerning trend.

SOURCE     Ars Technica: AI (tier 1)
   ↓
DOCUMENT   dd17b04d-b1ad-44ab-b2da-4f0a252bd597
           https://arstechnica.com/tech-policy/2026/10/openai-google-pushing-new-ai-models-easily-tricked-into-making-nudes-researchers-say
   ↓
EVIDENCE   2 extracted excerpts
   ↓
MODEL RUN  4 runs, methodology 1.0.1
   ↓
SCORE      -79  (Strongly adverse)
   ↓
CONFIDENCE 80%