White Circle says RISE jailbreak bypasses image safety filters in OpenAI and Google models
Overview
White Circle researchers say off-the-shelf tools bypassed safety filters on DALL·E 3, Google's Nano Banana 2 and GPT-Image-2 about 13 percent of the time, producing nude images.
The researchers also say ChatGPT Images 2.5 and Nano Banana 2.1 are significantly easier to break than their predecessors. Neither OpenAI nor Google responded to Ars Technica's requests for comment. The researchers argue that mandatory independent safety testing is needed.
Written by AI from the articles below · updated Oct 14, 7:41 AM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
- Ars Technica · AINewsWhite Circle says OpenAI and Google image models are easily jailbroken into making nudes
AIWhite Circle researchers say off-the-shelf tools bypassed safety filters on DALL·E 3, Nano Banana 2 and GPT-Image-2 about 13 percent of the time. They also say ChatGPT Images 2.5 and Nano Banana 2.1 are significantly more breakable than their predecessors. Neither OpenAI nor Google responded to Ars Technica's requests for comment, and the researchers argue that mandatory independent safety testing is needed.
Heat trend
Not enough continuous observations to show a trend yet.