Researchers say off-the-shelf tools bypass OpenAI, Google image filters
White Circle, a safety research group backed by people from Anthropic, DeepMind, Hugging Face and OpenAI, published a preprint describing a RISE attack pipeline that it says bypassed image filters on DALL·E 3, Google's Nano Banana 2 and GPT-Image-2 using off-the-shelf tools and API calls. Ars Technica reported that about 13 percent of attack prompts on DALL·E 3 produced human-verified explicit images, under 10 percent for Nano Banana 2 and about 3 percent for GPT-Image-2. White Circle plans to open-source the pipeline and present the work at NeurIPS 2026; TechNews also reported, without corroboration, that newer ChatGPT Images 2.5 and Nano Banana 2.1 were easier to bypass.
The researchers argue that the defenses built by leading labs are far weaker than the labs acknowledge, and they call for independent, third-party safety evaluation rather than self-assessment by the companies.
Story text
Size Width * Links
* Subscribers only Learn more
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.
#White Circle#OpenAI#Google#DALL·E 3#GPT-Image-2#Nano Banana 2#ChatGPT Images 2.5#Nano Banana 2.1#NeurIPS 2026