gpt-imageIMAGE
RISE 攻击测试：新图像模型更易被越狱
对前沿图像模型做 RISE 攻击测试，GPT-Image-2.5-Flash 攻击成功率涨到 2.4 倍，Nano Banana 2.1 涨 2.2 倍。
White Circle
@whitecircle13 天前收录原推发布于 2026.10.15
White Circle @whitecircle原推发布于 2026.10.15
We tested RISE on frontier image models, and it turns out that newer models are easier to break. Successful attacks went up 2.4× on GPT-Image-2.5-Flash and 2.2× on Nano Banana 2.1.
x.com/whitecircle/status/2110420705720316394采集于 2026.10.15 10:21