# White Circle finds new OpenAI and Google image models can be tricked into nudes

> Researchers say the latest versions of DALL·E 3, ChatGPT Images 2.5, Nano Banana 2.1 and GPT‑Image‑2 generate explicit images about 13% of the time when using a simple off‑the‑shelf jailbreak.

Oossa · 2026-10-14 · https://oossa.com/en/white-circle-finds-new-openai-and-google-image-models-can-be-tricked-into-nudes

White Circle, a safety‑focused public‑benefit corp, released a preprint showing that its RISE attack can bypass the safety filters of OpenAI’s ChatGPT Images 2.5, Google’s Nano Banana 2.1, DALL·E 3 and GPT‑Image‑2. Using only publicly available tools, the method generated non‑consensual sexual images in roughly 13 % of attempts. The researchers say the newer models are “significantly more breakable” than earlier versions. OpenAI and Google have not commented on the findings.

## The facts

- Bypass rate: about 13 % of prompts produced explicit images.
- Models tested: DALL·E 3, GPT‑Image‑2, ChatGPT Images 2.5 and Nano Banana 2.1.

## Why it matters

If the filters remain weak, anyone can use these popular image generators to create harmful nude content with minimal technical skill.

## Sources & references

1. [OpenAI, Google pushing new AI models easily tricked into making nudes, researchers say](https://arstechnica.com/tech-policy/2026/10/openai-google-pushing-new-ai-models-easily-tricked-into-making-nudes-researchers-say/) – Ars Technica, 2026-10-14

Last updated: 2026-10-14