What happened
White Circle researchers found that Google's Nano Banana 2 and OpenAI's DALL·E 3 and GPT-Image-2 could be bypassed into generating explicit images about 13 percent of the time. Testing showed the newer ChatGPT Images 2.5 and Nano Banana 2.1 are significantly more breakable.
Why it matters
White Circle said this bypass rate is the highest of any method tested under the same standard. The group is positioning itself as an external evaluator of frontier labs.
What to watch
White Circle will share the research at NeurIPS 2026 this November.
WHO IT HITSAI safety teams at Google and OpenAI face external pressure to harden image-generation safeguards. Policy makers weighing mandatory third-party safety testing now have a concrete bypass rate to cite.
Summaries like this, in your inbox every morning.
White Circle's findings come as AI-generated explicit content is spreading unchecked. A Google co-authored study this July found that 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. Google researchers concluded the most critical intervention point remains at the source.
Google's research also found that 20 percent of online users who admitted to making AI nudes did so "for fun" or "as a joke." Nearly a third were merely "experimenting with tech." In response, the Internet Watch Foundation warned last week that urgent action is needed after finding 6,310 AI images meeting the legal definition of CSAM in the first six months of 2026.
White Circle's Shilov argues frontier labs won't invest in safety testing until it's mandatory. The group is in early talks with government officials about third-party safety testing.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Anthropic co-founder Jack Clark told a University of Tokyo audience on October 13 that the chance of an AI str…
At a press briefing for Anthropic Japan's first anniversary, co-founder Jack Clark said on October 14 that AI…
Americans for Responsible Innovation, former Rep
A Harvard survey of more than 5,400 Americans found 53% of men said they knew of Susan Milner and Craig Dalton…
Tech Against Terrorism's Counter-Terrorism AI Benchmark tested 134 AI models with more than 200 terrorist-styl…
While testing a natural-language AI agent on BigQuery, the developer found answers included data from tables t…
